Product recommendation method and device, equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a product recommendation method and device, equipment and a storage medium, and the method comprises the steps: obtaining the purchase demand information of a target user; matching a corresponding target historical user model from a user model library according to the purchase demand information; correspondingly extracting candidate insurance products of historical users according to the target historical user model; marking a first product keyword for the candidate insurance product according to the purchase demand information; obtaining a second product keyword selected by the target user from the first product keywords; and adopting the matching model, screening out a target insurance product from the candidate insurance products according to the second product keyword, and recommending the target insurance product to the target user. The method is applied to financial science and technology business processing scenes, the problem that an existing product recommendation method is difficult to accurately adapt to customer requirements is solved, and the product recommendation accuracy and the user satisfaction degree are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, applied to financial technology business processing scenarios, and in particular to a product recommendation method, device, equipment and storage medium. Background Art

[0002] The rapid development of the insurance industry has driven the complexity and diversification of the insurance market, providing consumers with a wider range of insurance product choices. However, the continuous upgrading and iteration of the insurance market has led to information overload for consumers when purchasing insurance. The large amount of complex insurance information makes it difficult for customers to effectively sift through and make informed decisions. In addition, high premiums, misunderstandings of insurance terms, and improper assessment of potential risks further increase customers' purchasing difficulties.

[0003] The traditional insurance sales model relies heavily on the personal experience and advice of insurance salespeople or agents, leading customers to passively accept information. Alternatively, some consumers simply follow recommendations from friends and family and purchase insurance products. These practices often overlook individual circumstances and needs, resulting in a mismatch between the selected insurance products and actual needs. In the digital age, insurance companies generally underutilize the vast amounts of user data in their databases, making it difficult to extract valuable insights and develop recommendations tailored to customer needs.

[0004] In view of this, there is an urgent need for a product recommendation method that can accurately adapt to customer needs to improve customer satisfaction. Summary of the Invention

[0005] The present invention provides a product recommendation method, device, equipment and storage medium to solve the problem that existing product recommendation methods are difficult to accurately adapt to customer needs.

[0006] In a first aspect, a product recommendation method is provided, comprising:

[0007] Obtain the purchase demand information entered by the target user;

[0008] According to the purchase demand information, matching the corresponding target historical user model from the user model library;

[0009] Extracting candidate insurance products for a number of historical users according to the matched target historical user models;

[0010] labeling a first product keyword for a plurality of candidate insurance products according to the purchase demand information;

[0011] Obtaining a second product keyword selected by the target user from the first product keyword;

[0012] The trained matching model is used to screen out a target insurance product from a number of candidate insurance products according to the second product keyword, and the target insurance product is recommended to the target user.

[0013] In a second aspect, a product recommendation device is provided, comprising:

[0014] Demand acquisition module, used to obtain the purchase demand information input by the target user;

[0015] A model matching module is used to match the corresponding target historical user model from the user model library according to the purchase demand information;

[0016] A product extraction module is used to extract candidate insurance products for a number of historical users according to the matched target historical user models;

[0017] a keyword tagging module, configured to tag a first product keyword for a plurality of candidate insurance products according to the purchase demand information;

[0018] A keyword selection module, configured to obtain a second product keyword selected by the target user from the first product keyword;

[0019] The product matching module is used to use the trained matching model to screen out a target insurance product from a plurality of candidate insurance products according to the second product keyword, and recommend the target insurance product to the target user.

[0020] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned product recommendation method when executing the computer program.

[0021] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned product recommendation method are implemented.

[0022] In the solution implemented by the above-mentioned product recommendation method, device, computer equipment and storage medium, by obtaining the purchase demand information input by the target user and matching the target historical user model from the user model library according to the purchase demand information, it is possible to quickly find a user group similar to the target user, thereby shortening the recommendation process; by extracting candidate insurance products from multiple target historical user models based on the matched several target historical user models, it is possible to ensure that the insurance products to be recommended are diverse, and because the candidate insurance products are extracted based on the historical user model matching the target user, the products are guaranteed to have a high relevance; by labeling the first product keywords for several candidate insurance products according to the purchase demand information, the first product keywords are made more in line with the needs and concerns of the target user; by providing a user word selection mechanism, the second product keywords selected by the target user from the first product keywords are obtained, further improving the precise adaptation of the second product keywords to user needs; by adopting a matching model, the target insurance product is screened out from several candidate insurance products according to the second product keyword and recommended to the target user, thereby realizing intelligent and personalized product matching and recommendation, and improving the accuracy of product recommendations and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0024] Figure 1 is an exemplary system architecture diagram of a product recommendation method according to an embodiment of the present invention;

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

[0026] Figure 3 yes Figure 1 A schematic flow chart of a specific implementation of step S40;

[0027] Figure 4 yes Figure 1 A schematic flow chart of a specific implementation method before step S20;

[0028] Figure 5 is a structural diagram of a product recommendation device in one embodiment of the present invention;

[0029] Figure 6 It is a structural diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of the application are for the purpose of describing specific embodiments only and are not intended to limit this application. The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0031] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0032] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0033] like Figure 1 As shown, system architecture 100 may include a terminal device 101, a network 102, and a server 103. Terminal device 101 may be a laptop computer 1011, a tablet computer 1012, or a mobile phone 1013. Network 102 is a medium for providing a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0034] The user can use the terminal device 101 to interact with the server 103 via the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0035] The terminal device 101 can be various electronic devices with a display screen and supporting web browsing. In addition to the laptop computer 1011, tablet computer 1012 or mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV) player, a laptop computer and a desktop computer, etc.

[0036] The server 103 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal device 101 .

[0037] It should be noted that the product recommendation method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the product recommendation device is generally set in the server / terminal device.

[0038] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0039] Continue reading Figure 2 As shown, Figure 2 A flowchart of a product recommendation method provided by an embodiment of the present invention includes the following steps:

[0040] S20: Obtaining the purchase demand information input by the target user;

[0041] The product recommendation method provided by the present invention can be applied to intelligent product recommendation engines in various application scenarios. The intelligent product recommendation engine is usually implemented by a server, which is connected to a client via a network. The server can obtain the purchase demand information input by the target user through the client, wherein the client may include but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices. The server can be implemented with an independent server or a server cluster composed of multiple servers. The present invention is described in detail below through specific embodiments.

[0042] Specifically, the purchase demand information input by the target user can be obtained. For example, a product recommendation application can be pre-installed on the client. This application provides a user input interface that allows the target user to manually enter or select one or more purchase demand information, or calls a client API to retrieve purchase demand information collected by a storage unit or other application on the client. The target user refers to the individual or group currently in need of product recommendation services. Their attributes are described by both static characteristics and dynamic behaviors. For example, static characteristics include age, gender, occupation, region, and risk preference; dynamic behaviors include historical interaction data (such as product click records, policy comparison behavior, and keyword search history), purchase paths, and real-time feedback data (such as ignores, clicks, and duration of stay). Purchase demand information refers to the specific content used to describe the user's insurance needs or preferences, including explicit and implicit needs. Explicit needs are product needs explicitly expressed by users; implicit needs are product needs derived from user information mining using artificial intelligence or knowledge graphs. For example, a user's mention of "ICU coverage" may imply a need for high-end medical services. Purchase demand information may include but is not limited to budget range, insurance type, insurance amount requirement, claim settlement method, coverage scope, insurance benefits, etc.

[0043] For example, the target user enters a sentence on the user input interface, "I want to buy a medical insurance that covers outpatient and hospitalization, with a budget of less than 500 yuan." Natural language processing (NLP) technology is used to parse the text, such as entity recognition and intent classification, to extract the user's purchasing demand information, including product type - medical insurance, coverage - outpatient and medical care, and budget range - less than 500 yuan.

[0044] S30: Matching a corresponding target historical user model from a user model library according to the purchase demand information;

[0045] Specifically, a user model library is pre-built, containing historical user models of multiple historical users. Each historical user model is composed of historical demand characteristics, including historical budget range, historical insurance type, historical coverage requirements, historical claims settlement methods, historical coverage scope, and other characteristics. Based on the target user's purchasing demand information, a corresponding target historical user model is matched from the user model library. For example, the similarity between the target user's purchasing demand information and each historical user model can be calculated using cosine similarity or a machine learning model. The top-N similar historical user models are then screened out as the target historical user model.

[0046] For example, when the target user's purchase demand information indicates that the target user needs medical insurance, based on the purchase demand information, cosine similarity or graph neural network (such as LightGCN) is used to find the top-N historical user models that have purchased similar insurance types from the user model library.

[0047] In some embodiments, step S30, i.e., matching a corresponding target historical user model from a user model library based on the purchase demand information, may include the following steps:

[0048] extracting key demand features based on the purchase demand information;

[0049] Specifically, based on the target user's purchasing demand information, demand keywords are extracted through word segmentation, stop word removal, entity recognition and other technologies, and then word embedding models (such as Word2Vec, BERT) or TF-IDF are used to convert demand keywords into numerical vectors, which are the key demand features.

[0050] Acquire a plurality of pre-built historical user models from a user model library, wherein the historical user models include historical demand characteristics of historical users;

[0051] Specifically, several pre-built historical user models are obtained from the user model library. For example, index words are extracted based on purchase demand information, and several historical user models are retrieved from the user model library using the index words. Each historical user model is stored in the form of a vector, and the historical user model contains the historical demand characteristics of the historical user.

[0052] Inputting the key demand feature and a plurality of the historical demand features into an analogy model for feature matching in sequence, and calculating the feature similarity between the key demand feature and the historical demand features;

[0053] Specifically, the target user's key demand features and the historical demand features of several historical users are sequentially input into an analogy model for feature matching. This analogy model can be composed of a neural network, support vector machine, or twin network architecture. Using a classification model, the feature similarity between the key demand features and each historical demand feature is calculated. For example, the key demand features and each historical demand feature are combined into feature pairs. Each feature pair is sequentially input into the twin network architecture of the classification model. After generating an embedding vector through a shared weight network, the cosine similarity is calculated to obtain the feature similarity between the key demand features and each historical demand feature.

[0054] The historical user models are sorted based on the feature similarity to obtain a model list, and a preset number of historical user models with the highest order are extracted from the model list to be determined as target historical user models.

[0055] Specifically, use quick sort or heap sort to sort each historical user model by feature similarity to obtain a model list. Obtain a preset number of models, such as Top-N or Top-n%, and extract the preset number of historical user models that are ranked at the top of the model list and determine them as target historical user models.

[0056] The above solution accurately captures the core needs of target users by extracting key demand features from purchase demand information. By obtaining historical user models from the user model library and calculating feature similarity using an analogy model, the experience of historical users can be leveraged to provide more accurate recommendations or services for the current target user. By calculating the similarity between key demand features and historical demand features and sorting historical user models, the degree of match between different user models and the current user's needs can be quantified, thereby selecting the historical user model that best matches the target user's purchase needs, improving product recommendation accuracy and user satisfaction.

[0057] S40: extracting candidate insurance products for a number of historical users according to the matched target historical user models;

[0058] Specifically, for each matched target historical user model, all historical insurance products associated with each target historical user model are extracted as candidate insurance products. For example, if historical user A purchased insurance products X and Y, and historical user B purchased insurance products Y and Z, then the candidate insurance products are X, Y, and Z.

[0059] In some embodiments, all historical insurance products that a historical user has purchased may cover a variety of types and needs. In order to further screen out candidate insurance products that better meet the needs of the target user, it is necessary to further consider the insurance needs of the target user. Figure 3 Step S40, i.e., extracting candidate insurance products of several historical users according to the matched target historical user models, may include the following steps S401-S402:

[0060] S401: Acquire all historical insurance products associated with the target historical user model;

[0061] S402: Acquire product information of the historical insurance product, match the product information with the purchase demand information, and screen to obtain candidate insurance products that match the purchase demand information.

[0062] Specifically, the target user's historical purchase records are obtained, and all insurance products previously purchased by the historical user are extracted from these historical purchase records. Product information for these historical insurance products is obtained, which may include premium price, primary insurance type, primary claim settlement method, primary insurance benefit, etc. This product information is matched with the target user's purchase requirement information, and historical insurance products that match this purchase requirement information are screened as candidate insurance products.

[0063] In some embodiments, the above step S402 may include one or more of the following steps:

[0064] Matching the first insurance type in the product information with the second insurance type in the purchase demand information, and if they match, determining the historical insurance product corresponding to the first insurance type as a candidate insurance product;

[0065] Specifically, when matching insurance types, the similarity between the first insurance type in the product information and the second insurance type in the purchase demand information is calculated based on the matching method of classification attributes. If the similarity meets the first preset threshold, it is determined that the first insurance type and the second insurance type match, and the historical insurance product corresponding to the first insurance type is determined as a candidate insurance product.

[0066] Matching the premium price in the product information with the budget range in the purchase demand information; if they match, determining the historical insurance product corresponding to the premium price as a candidate insurance product;

[0067] Specifically, when matching premium prices, the premium price in the product information is matched with the budget range in the purchase demand information based on the matching method of numerical attributes. If the premium price meets the budget range, the premium price is indeed matched with the budget range, and the historical insurance product corresponding to the premium price is determined as a candidate insurance product.

[0068] Matching the first claim settlement method in the product information with the second claim settlement method in the purchase demand information; if they match, determining the historical insurance product corresponding to the first claim settlement method as a candidate insurance product;

[0069] Specifically, when matching the claim methods, the similarity between the first claim method in the product information and the second claim method in the purchase demand information is calculated based on the matching method of the classification attributes. If the similarity meets the second preset threshold, it is determined that the first claim method and the second claim method match, and the historical insurance product corresponding to the first claim method is determined as a candidate insurance product.

[0070] The first insurance benefit in the product information is matched with the second insurance benefit in the purchase demand information. If they match, the historical insurance product corresponding to the first insurance benefit is determined as a candidate insurance product.

[0071] Specifically, when matching insurance benefits, the similarity between the first insurance benefit in the product information and the second insurance benefit in the purchase demand information is calculated based on the matching method of classification attributes. If the similarity meets the third preset threshold, it is determined that the first insurance benefit and the second insurance benefit match, and the historical insurance product corresponding to the first insurance benefit is determined as a candidate insurance product.

[0072] S50: labeling a first product keyword for a plurality of candidate insurance products according to the purchase demand information;

[0073] Specifically, product information for several candidate insurance products is obtained. Based on the target user's purchasing needs and product information, the first product keyword for each candidate insurance product is determined and labeled. For example, the product information for the candidate insurance products is segmented, stop words are removed, and entity recognition is performed to extract product keywords. The extracted product keywords are then matched with the purchasing needs information, and matching product keywords are selected and labeled as the first product keyword for the candidate insurance product.

[0074] For example, a target user's purchasing need information might be, "I want to purchase medical insurance with outpatient coverage, and my budget is under 500 yuan." The product information for a candidate insurance product might be, "Insurance Name: Major Illness Medical Insurance; Coverage: Outpatient and Hospitalization; Premium: 400 yuan / year." Based on this purchasing need information, the first product keyword for this candidate insurance product would be "medical insurance, outpatient coverage, premium 400 yuan."

[0075] In some embodiments, when there is a large amount of complex product information for candidate insurance products, directly matching the target user's purchasing needs with the complex product information makes it difficult to ensure that the matched product keywords can represent the product characteristics of the product. Specifically, step S50, i.e., labeling the first product keyword for several candidate insurance products based on the purchasing need information, may include the following steps:

[0076] extracting product keywords based on the product information of the candidate insurance product;

[0077] Specifically, product information for candidate insurance products is segmented, stop words are removed, and entity recognition is performed to extract product keywords. For example, for text data such as product descriptions and terms and conditions documents, NLP techniques are used for word segmentation, part-of-speech tagging, and named entity recognition. For structured data such as coverage in a table, field values are directly extracted (e.g., "hospitalization allowance: 300 yuan per day").

[0078] Calculating the importance weights of the product keywords in the product information, and selecting a preset number of product keywords with higher importance weights as comparison product keywords;

[0079] Specifically, the importance weight of each product keyword in the product information is calculated. For example, the importance weight of the product keyword is calculated based on the frequency of occurrence, location, whether it belongs to the core guarantee clause, etc. The product keywords are sorted in descending order by importance weight, and a preset number (N) of product keywords with the highest importance weight are selected as comparison product keywords for subsequent comparison with the purchase demand information.

[0080] For example, we use the TF-IDF algorithm to evaluate term frequency and inverse document frequency, or we use the knowledge graph to query the insurance-related strength of specialized terms and assign them higher weights. Furthermore, we determine the weight of product keywords based on their position within the product information: keywords in the title are weighted x1.5, and keywords in the first paragraph are weighted x1.2.

[0081] Sequentially calculating the keyword similarity between the compared product keywords and the demand keywords in the purchase demand information;

[0082] Specifically, demand keywords are extracted from the purchase demand information, and the keyword similarity between each comparison product keyword and the demand keyword is calculated in sequence. For example, cosine similarity, edit distance, or pre-trained word vector models (such as Word2Vec and BERT) are used to calculate the semantic similarity between the comparison product keywords and the demand keywords.

[0083] The keyword similarity is compared with a preset similarity threshold, and the matched product keyword that is greater than or equal to the similarity threshold is determined as the first product keyword of the candidate insurance product.

[0084] Specifically, a preset similarity threshold is obtained, and the keyword similarity between each comparison product keyword and the demand keyword is compared with the similarity threshold. Comparison product keywords with a similarity greater than or equal to the similarity threshold are determined as the first product keyword for the candidate insurance product. The similarity threshold can be determined based on historical matching data or set by professionals.

[0085] The above solution extracts product keywords and calculates their importance weights, achieving information dimensionality reduction and focusing on product keywords that best represent product characteristics. By selecting a preset number of product keywords with the highest importance weights as comparison keywords and calculating their similarity with the demand keywords in the purchase demand information, we ensure that the first matched product keyword not only represents the core characteristics of the candidate insurance product but also closely matches the user's purchasing needs, thereby improving matching accuracy and user satisfaction.

[0086] S60: Acquire a second product keyword selected by the target user from the first product keywords;

[0087] Specifically, a user-selected keyword mechanism is set up to aggregate the first product keywords of each candidate insurance product, convert them into natural language, and present them to the target user for selection. After the user selects a second product keyword of greater interest from among the multiple first product keywords, the second product keyword is obtained.

[0088] S70: Using the trained matching model, based on the second product keyword, a target insurance product is screened out from a plurality of candidate insurance products, and the target insurance product is recommended to the target user.

[0089] Specifically, the second product keywords and several candidate insurance products are input into the trained matching model in sequence, and the matching model is used to encode the second product keywords and the candidate insurance products into semantic vectors respectively. The matching scores between the encoded second product keywords and the candidate insurance products are calculated, and the target insurance product is determined from the candidate insurance products based on the matching scores, and the target insurance product is recommended to the target user.

[0090] In some embodiments, step S70, i.e., using the trained matching model to select a target insurance product from the plurality of candidate insurance products based on the second product keyword, may include the following steps:

[0091] Inputting the candidate insurance product and the second product keyword into the trained matching model, extracting a candidate product feature vector based on the candidate insurance product, and extracting a keyword feature vector based on the second product keyword;

[0092] Specifically, the candidate insurance product and the second product keywords are input into the trained matching model, where the candidate insurance product may be composed of textual information (such as name, terms, coverage, product description, etc.) and numerical information (such as premium price, deductible, etc.). The textual information is converted into a high-dimensional dense text vector through a multimodal encoder, and the numerical information is normalized and concatenated with the text vector to obtain the candidate product feature vector; the second product keywords are semantically analyzed to capture the deep semantics of the user's intention and generate a keyword feature vector.

[0093] It should be noted that the matching model can adopt a dual-tower heterogeneous network architecture, and the candidate product feature vectors and keyword feature vectors are generated through independent sub-networks respectively to ensure the consistency of the feature space.

[0094] Calculating a first matching score between the candidate product feature vector and the keyword feature vector, and sorting the candidate insurance products by size according to the first matching score to obtain a product list;

[0095] Specifically, based on the candidate product feature vectors and keyword feature vectors, a first match score is calculated between the candidate product feature vectors and the keyword feature vectors. For example, the matching score calculation process uses a multi-dimensional matching strategy. The basic matching layer calculates the vector dot product as the initial similarity. The interactive matching layer uses the Cross-Attention mechanism to capture the fine-grained correspondence between the candidate product feature vectors and the keyword feature vectors. The fusion layer then weights and combines the two matching results (i.e., the initial similarity and the fine-grained correspondence) to generate the final first match score. Afterwards, the candidate insurance products are sorted in descending order according to the size of the first match score to obtain a product list.

[0096] A preset number of candidate insurance products ranked high in the product list are selected and determined as target insurance products.

[0097] Specifically, a preset number, such as Top-N or Top-n%, is obtained, and a preset number of candidate insurance products with the highest ranking are extracted from the product list and determined as target insurance products.

[0098] In some embodiments, the process of training the matching model may also be included. Figure 4 Before step S20, that is, before obtaining the purchase demand information input by the target user, the following steps S10-S14 may also be included:

[0099] S10: Acquire a pre-collected training data set, wherein the training data set includes historical demand keywords, historical insurance products, and actual matching relationships between the historical demand keywords and the historical insurance products;

[0100] Specifically, a pre-collected training data set is obtained, which includes historical demand keywords, historical insurance products, and the actual matching relationship between historical demand keywords and historical insurance products. The actual matching relationship is used as a supervisory signal in the training process to calculate losses.

[0101] S20: Inputting the training data set into the matching model to be trained in sequence, extracting the historical demand feature vector of the historical demand keyword and the historical product feature vector of the historical insurance product;

[0102] Specifically, the training data set is input into the matching model to be trained in sequence. The matching model can adopt a dual-tower heterogeneous network architecture, which includes a demand tower and a product tower. The pre-trained language model is used through the demand tower to extract the historical demand feature vector according to the input historical demand keywords; and the hybrid coding structure is used through the product tower to extract the historical product feature vector according to the input historical insurance products (including text information and numerical information).

[0103] S30: Calculating a second matching score between the historical demand feature vector and the historical product feature vector, and determining a matching relationship for training according to the second matching score;

[0104] Specifically, a second matching score is calculated between the historical demand feature vector and the historical product feature vector. For example, a basic matching layer calculates the vector dot product as a preliminary similarity based on the historical demand feature vector and the historical product feature vector. The interactive matching layer uses a Cross-Attention mechanism to capture the fine-grained correspondence between the historical demand feature vector and the historical product feature vector. The fusion layer weights and combines the two matching results (i.e., the preliminary similarity and the fine-grained correspondence) to generate a second matching score. A matching relationship for training is determined based on the second matching score. For example, when the second matching score is greater than or equal to a preset threshold, the matching relationship for training is determined to be a match; when the second matching score is less than the preset threshold, the matching relationship for training is determined to be a mismatch.

[0105] S40: Obtain a preset loss function, and update model parameters based on the matching relationship for training, the actual matching relationship, and the loss function;

[0106] S50: Iteratively train the matching model after the model parameters are updated until the model converges to obtain a trained matching model.

[0107] Specifically, a preset loss function is obtained, and a loss value is calculated by combining the training matching relationship, the actual matching relationship, and the loss function. The gradient is calculated based on the loss value and the model parameters are updated. The loss function can use contrastive loss, triple loss, or cross entropy, etc. Following the above steps, the matching model with updated model parameters is iteratively trained using the training dataset until the model converges, resulting in a trained matching model.

[0108] Through the above solution, through automated feature extraction, matching relationship learning and parameter optimization, an efficient and accurate insurance demand-product matching model is realized, effectively improving product matching accuracy.

[0109] The product recommendation method provided by the embodiment of the present invention obtains the purchase demand information input by the target user, matches the target historical user model from the user model library according to the purchase demand information, and can quickly find user groups similar to the target user, thereby shortening the recommendation process; by extracting candidate insurance products from multiple target historical user models based on the matched several target historical user models, it can be ensured that the insurance products to be recommended are diverse, and because the candidate insurance products are extracted based on the historical user model matching the target user, it is guaranteed that the products have high relevance; by labeling the first product keywords for several candidate insurance products according to the purchase demand information, the first product keywords are made more in line with the needs and concerns of the target user; by providing a user word selection mechanism, the second product keywords selected by the target user from the first product keywords are obtained, thereby further improving the accurate adaptation of the second product keywords to the user needs; by adopting a matching model, the target insurance product is screened out from several candidate insurance products according to the second product keyword, and recommended to the target user, thereby realizing intelligent and personalized product matching and recommendation, and improving the accuracy of product recommendation and user satisfaction.

[0110] It should be understood that the order of execution of the steps in the above embodiments does not necessarily mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0111] In one embodiment, a product recommendation device is provided, which corresponds to the product recommendation method in the above embodiment. Figure 5 As shown, the product recommendation device includes a demand acquisition module 201, a model matching module 202, a product extraction module 203, a keyword annotation module 204, a keyword selection module 205 and a product matching module 206. The functional modules are described in detail as follows:

[0112] Demand acquisition module 201, used to obtain the purchase demand information input by the target user;

[0113] Model matching module 202, used to match the corresponding target historical user model from the user model library according to the purchase demand information;

[0114] A product extraction module 203 is configured to extract candidate insurance products corresponding to a plurality of historical users according to the matched plurality of target historical user models;

[0115] A keyword tagging module 204 is configured to tag a first product keyword for a plurality of candidate insurance products according to the purchase demand information;

[0116] A keyword selection module 205 is configured to obtain a second product keyword selected by the target user from the first product keywords;

[0117] The product matching module 206 is configured to use the trained matching model to select a target insurance product from a plurality of candidate insurance products according to the second product keyword, and recommend the target insurance product to the target user.

[0118] In one embodiment, the model matching module 202 is specifically configured to:

[0119] extracting key demand features based on the purchase demand information;

[0120] Acquire a plurality of pre-built historical user models from a user model library, wherein the historical user models include historical demand characteristics of historical users;

[0121] Inputting the key demand feature and a plurality of the historical demand features into an analogy model for feature matching in sequence, and calculating the feature similarity between the key demand feature and the historical demand features;

[0122] The historical user models are sorted based on the feature similarity to obtain a model list, and a preset number of historical user models with the highest order are extracted from the model list to be determined as target historical user models.

[0123] In one embodiment, the product extraction module 203 is specifically configured to:

[0124] Obtain all historical insurance products associated with the target historical user model;

[0125] The product information of the historical insurance products is obtained, the product information is matched with the purchase demand information, and candidate insurance products that match the purchase demand information are screened.

[0126] In one embodiment, the product extraction module 203 is further configured to:

[0127] Matching the first insurance type in the product information with the second insurance type in the purchase demand information, and if they match, determining the historical insurance product corresponding to the first insurance type as a candidate insurance product;

[0128] Matching the premium price in the product information with the budget range in the purchase demand information; if they match, determining the historical insurance product corresponding to the premium price as a candidate insurance product;

[0129] Matching the first claim settlement method in the product information with the second claim settlement method in the purchase demand information; if they match, determining the historical insurance product corresponding to the first claim settlement method as a candidate insurance product;

[0130] The first insurance benefit in the product information is matched with the second insurance benefit in the purchase demand information. If they match, the historical insurance product corresponding to the first insurance benefit is determined as a candidate insurance product.

[0131] In one embodiment, the keyword tagging module 204 is specifically configured to:

[0132] extracting product keywords based on the product information of the candidate insurance product;

[0133] Calculating the importance weights of the product keywords in the product information, and selecting a preset number of product keywords with higher importance weights as comparison product keywords;

[0134] Sequentially calculating the keyword similarity between the compared product keywords and the demand keywords in the purchase demand information;

[0135] The keyword similarity is compared with a preset similarity threshold, and the matched product keyword that is greater than or equal to the similarity threshold is determined as the first product keyword of the candidate insurance product.

[0136] In one embodiment, the product matching module 206 is specifically configured to:

[0137] Inputting the candidate insurance product and the second product keyword into the trained matching model, extracting a candidate product feature vector based on the candidate insurance product, and extracting a keyword feature vector based on the second product keyword;

[0138] Calculating a first matching score between the candidate product feature vector and the keyword feature vector, and sorting the candidate insurance products by size according to the first matching score to obtain a product list;

[0139] A preset number of candidate insurance products ranked high in the product list are selected and determined as target insurance products.

[0140] In one embodiment, the product recommendation device further includes a model training module, specifically configured to:

[0141] Acquire a pre-collected training data set, the training data set including historical demand keywords, historical insurance products, and actual matching relationships between the historical demand keywords and the historical insurance products;

[0142] Inputting the training data set into the matching model to be trained in sequence, extracting the historical demand feature vector of the historical demand keyword and the historical product feature vector of the historical insurance product;

[0143] Calculating a second matching score between the historical demand feature vector and the historical product feature vector, and determining a matching relationship for training according to the second matching score;

[0144] Obtaining a preset loss function, and updating model parameters based on the matching relationship for training, the actual matching relationship, and the loss function;

[0145] The matching model after the model parameters are updated is iteratively trained until the model converges to obtain the trained matching model.

[0146] The present invention provides a product recommendation device, which obtains purchase demand information input by a target user and matches a target historical user model from a user model library according to the purchase demand information, thereby quickly finding a user group similar to the target user, thereby shortening the recommendation process; by extracting candidate insurance products from multiple target historical user models based on several matched target historical user models, it can be ensured that the insurance products to be recommended are diverse, and because the candidate insurance products are extracted based on the historical user model that matches the target user, it is guaranteed that the products have high relevance; by labeling several candidate insurance products with first product keywords according to the purchase demand information, the first product keywords are made more in line with the target user's demand concerns; by providing a user word selection mechanism, the second product keywords selected by the target user from the first product keywords are obtained, thereby further improving the accurate adaptation of the second product keywords to the user's needs; by adopting a matching model, the target insurance product is screened out from several candidate insurance products according to the second product keyword, and recommended to the target user, thereby realizing intelligent and personalized product matching and recommendation, and improving the accuracy of product recommendation and user satisfaction.

[0147] For the specific definition of the product recommendation device, please refer to the definition of the product recommendation method above and will not be repeated here. Each module in the above-mentioned product recommendation device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0148] In order to solve the above technical problems, the present application also provides a computer device, please refer to Figure 6 , Figure 6 This is a basic structural block diagram of the computer device in this embodiment.

[0149] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 6 with a memory 61, a processor 62, and a network interface 63, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to a microprocessor, an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), an embedded device, etc.

[0150] The computer device may be a desktop computer, notebook computer, PDA, cloud server, etc. The computer device may interact with the user via a keyboard, mouse, remote control, touchpad, or voice control device.

[0151] The memory 61 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 61 can be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 can also be an external storage device of the computer device 6, such as a plug-in hard disk equipped on the computer device 6, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Of course, the memory 61 can also include both the internal storage unit of the computer device 6 and its external storage device. In this embodiment, the memory 61 is generally used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions of the product recommendation method. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or are to be output.

[0152] In some embodiments, the processor 62 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 62 is generally used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to execute computer-readable instructions or process data stored in the memory 61, such as computer-readable instructions for executing the product recommendation method.

[0153] The network interface 63 may include a wireless network interface or a wired network interface. The network interface 63 is generally used to establish a communication connection between the computer device 6 and other electronic devices.

[0154] The present application also provides another embodiment, namely, providing a computer-readable storage medium, which stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the product recommendation method as described above.

[0155] The computer device, computer-readable storage medium, and computer-readable instructions thereof provided in the embodiments of the present application obtain purchase demand information input by a target user through processor execution, match a target historical user model from a user model library based on the purchase demand information, and can quickly find a user group similar to the target user, thereby shortening the recommendation process; by extracting candidate insurance products from multiple target historical user models based on several matched target historical user models, it can be ensured that the insurance products to be recommended are diverse, and because the candidate insurance products are extracted based on the historical user models that match the target user, the products are guaranteed to have a high relevance; by labeling several candidate insurance products with first product keywords based on the purchase demand information, the first product keywords are made more in line with the needs and concerns of the target user; by providing a user word selection mechanism, the second product keywords selected by the target user from the first product keywords are obtained, thereby further improving the precise adaptation of the second product keywords to user needs; by adopting a matching model, the target insurance product is screened from several candidate insurance products based on the second product keyword and recommended to the target user, thereby realizing intelligent and personalized product matching and recommendation, and improving the accuracy of product recommendations and user satisfaction.

[0156] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0157] Obviously, the embodiments described above are only some of the embodiments of the present application, rather than all of the embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned specific embodiments, or to make equivalent replacements for some of the technical features therein. Any equivalent structure made using the contents of the present application specification and the accompanying drawings, directly or indirectly used in other related technical fields, is also within the scope of patent protection of the present application.

[0158] The non-Company software tools or components appearing in the embodiments of this application are merely examples and do not represent actual use.

Claims

1. A product recommendation method, characterized in that: include: Obtain the purchase demand information entered by the target user; According to the purchase demand information, matching the corresponding target historical user model from the user model library; Extracting candidate insurance products for a number of historical users according to the matched target historical user models; labeling a first product keyword for a plurality of candidate insurance products according to the purchase demand information; Obtaining a second product keyword selected by the target user from the first product keyword; The trained matching model is used to screen out a target insurance product from a number of candidate insurance products according to the second product keyword, and the target insurance product is recommended to the target user.

2. The method according to claim 1, wherein The step of matching a corresponding target historical user model from a user model library according to the purchase demand information includes: extracting key demand features based on the purchase demand information; Acquire a plurality of pre-built historical user models from a user model library, wherein the historical user models include historical demand characteristics of historical users; Inputting the key demand feature and a plurality of the historical demand features into an analogy model for feature matching in sequence, and calculating the feature similarity between the key demand feature and the historical demand features; The historical user models are sorted based on the feature similarity to obtain a model list, and a preset number of historical user models with the highest order are extracted from the model list to be determined as target historical user models.

3. The method according to claim 1, wherein Extracting candidate insurance products for a number of historical users according to the matched target historical user models includes: Obtain all historical insurance products associated with the target historical user model; The product information of the historical insurance products is obtained, the product information is matched with the purchase demand information, and candidate insurance products that match the purchase demand information are screened.

4. The method according to claim 3, wherein The matching of the product information with the purchase demand information to screen and obtain candidate insurance products that match the purchase demand information includes at least one of the following: Matching the first insurance type in the product information with the second insurance type in the purchase demand information, and if they match, determining the historical insurance product corresponding to the first insurance type as a candidate insurance product; Matching the premium price in the product information with the budget range in the purchase demand information; if they match, determining the historical insurance product corresponding to the premium price as a candidate insurance product; Matching the first claim settlement method in the product information with the second claim settlement method in the purchase demand information; if they match, determining the historical insurance product corresponding to the first claim settlement method as a candidate insurance product; The first insurance benefit in the product information is matched with the second insurance benefit in the purchase demand information. If they match, the historical insurance product corresponding to the first insurance benefit is determined as a candidate insurance product.

5. The method according to claim 1, wherein The step of labeling the first product keyword for the plurality of candidate insurance products according to the purchase demand information includes: extracting product keywords based on the product information of the candidate insurance product; Calculating the importance weights of the product keywords in the product information, and selecting a preset number of product keywords with higher importance weights as comparison product keywords; Sequentially calculating the keyword similarity between the compared product keywords and the demand keywords in the purchase demand information; The keyword similarity is compared with a preset similarity threshold, and the matched product keyword that is greater than or equal to the similarity threshold is determined as the first product keyword of the candidate insurance product.

6. The method according to claim 1, wherein The trained matching model is used to select a target insurance product from the plurality of candidate insurance products according to the second product keyword, including: Inputting the candidate insurance product and the second product keyword into the trained matching model, extracting a candidate product feature vector based on the candidate insurance product, and extracting a keyword feature vector based on the second product keyword; Calculating a first matching score between the candidate product feature vector and the keyword feature vector, and sorting the candidate insurance products by size according to the first matching score to obtain a product list; A preset number of candidate insurance products ranked high in the product list are selected and determined as target insurance products.

7. The method according to any one of claims 1 to 6, wherein Before obtaining the purchase demand information input by the target user, the method further includes: Acquire a pre-collected training data set, the training data set including historical demand keywords, historical insurance products, and actual matching relationships between the historical demand keywords and the historical insurance products; Inputting the training data set into the matching model to be trained in sequence, extracting the historical demand feature vector of the historical demand keyword and the historical product feature vector of the historical insurance product; Calculating a second matching score between the historical demand feature vector and the historical product feature vector, and determining a matching relationship for training according to the second matching score; Obtaining a preset loss function, and updating model parameters based on the matching relationship for training, the actual matching relationship, and the loss function; The matching model after the model parameters are updated is iteratively trained until the model converges to obtain the trained matching model.

8. A product recommendation device, characterized in that: include: Demand acquisition module, used to obtain the purchase demand information input by the target user; A model matching module is used to match the corresponding target historical user model from the user model library according to the purchase demand information; A product extraction module is used to extract candidate insurance products for a number of historical users according to the matched target historical user models; a keyword tagging module, configured to tag a first product keyword for a plurality of candidate insurance products according to the purchase demand information; A keyword selection module, configured to obtain a second product keyword selected by the target user from the first product keyword; The product matching module is used to use the trained matching model to screen out a target insurance product from a plurality of candidate insurance products according to the second product keyword, and recommend the target insurance product to the target user.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the product recommendation method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the product recommendation method according to any one of claims 1 to 7 are implemented.

Images