Session generation method and device for product information recommendation based on scoring model

Abstract

The invention discloses a product information recommendation session generation method and device based on a scoring model, a medium and equipment, relates to the field of artificial intelligence and the field of financial science and technology, and mainly aims to solve the problem that the accuracy of insurance product information recommended by an existing intelligent session is poor. Comprising the following steps: acquiring user communication information input by each application program, and calling associated information corresponding to the user communication information from an insurance data system; performing evaluation processing on the associated information based on a trained scoring model to obtain a recommendability score, the scoring model being obtained by performing model training on a deep neural network based on an operation sample; and determining a session keyword matched with the recommendable score, and generating a recommendation session of the product information based on the session keyword.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and financial technology, and in particular to a conversation generation method and device for product information recommendation based on a scoring model. Background Art

[0002] With the development of science and technology, obtaining data and using data modeling to solve corresponding problems has become a very common technical means. For example, various insurance systems will collect user data such as user product browsing history, and build a recommendation model for insurance products based on the collected data to recommend insurance products to users.

[0003] At present, when recommending existing insurance product information, usually after selecting the recommended insurance product, the fixed script text is called, and the conversation information of the insurance product is inserted into the script text to obtain the conversation content of the recommended insurance product for conversational recommendation. However, due to the single sentence of the fixed script text, the recommendation effect of insurance product information is poor, and the selected insurance products cannot meet the flexible recommendation needs of different users, which greatly reduces the accuracy of recommending insurance product information through intelligent conversation. Summary of the invention

[0004] In view of this, the present invention provides a method and device for generating a conversation for product information recommendation based on a scoring model, the main purpose of which is to solve the problem of poor accuracy of existing intelligent conversations for recommending insurance product information.

[0005] According to one aspect of the present invention, a method for generating a session for product information recommendation based on a scoring model is provided, comprising:

[0006] Obtaining user communication information entered by each application, and retrieving associated information corresponding to the user communication information from the insurance data system;

[0007] The associated information is evaluated based on a trained scoring model to obtain a recommendability score, wherein the scoring model is obtained by training a deep neural network based on the operation sample;

[0008] A conversation keyword matching the recommendable score is determined, and a recommendation conversation of product information is generated based on the conversation keyword.

[0009] Furthermore, before evaluating the associated information based on the trained scoring model to obtain a recommendability score, the method further includes:

[0010] Retrieving associated samples marked with different scores from the insurance data system and constructing a multi-input and multi-output deep neural network, wherein the associated samples include transaction operation samples, identity samples, and channel operation samples;

[0011] The deep neural network is trained based on the operation samples to obtain the scoring model.

[0012] Furthermore, the performing model training on the deep neural network based on the operation sample to obtain the scoring model includes:

[0013] When the deep neural network is trained based on the operation sample, the loss function is adjusted based on the communication value coefficient, and the scoring loss value is calculated based on the adjusted loss function. The communication value coefficient is used to characterize the evaluation results of transaction behaviors associated with user communication information.

[0014] Further, the recommendation session for generating product information based on the session keywords includes:

[0015] Retrieving a conversation template, wherein the conversation template is obtained by combining transaction speech sentences in advance;

[0016] The conversation keywords are inserted into the conversation template according to the word order and word semantics to obtain a recommended conversation.

[0017] Further, the determining of the conversation keywords matching the recommendable score includes:

[0018] Analyze the recommendation strength score, transaction expectation score, and conversation scenario score in the recommendable score;

[0019] Retrieve a preset scoring keyword mapping relationship, and query the conversation keywords that match the recommendable score based on the preset scoring keyword mapping relationship. The preset scoring keyword mapping relationship includes conversation keywords corresponding to different recommendation strength scores, transaction expectation scores, and conversation scenario scores. The conversation keywords are used to characterize product keywords and recommendation speech keywords of insurance products.

[0020] Furthermore, the acquiring of user communication information entered by each application program and retrieving associated information corresponding to the user communication information from the insurance data system includes:

[0021] Sending a communication information acquisition instruction to each of the applications, so that the application can feedback authorized user communication information;

[0022] In response to the feedback instruction of the application, sending the user communication information carried in the feedback instruction to the insurance data system, so that the insurance data system queries the associated information matching the user communication information for feedback;

[0023] The association information is received.

[0024] Furthermore, after generating a recommendation session of product information based on the session keywords, the method further includes:

[0025] Counting the product transaction results of the recommendation session;

[0026] The user communication information is marked based on the product transaction result to update the transaction operation information in the associated information.

[0027] According to another aspect of the present invention, a device for generating a conversation for recommending product information based on a scoring model is provided, comprising:

[0028] An acquisition module, used to acquire user communication information entered by each application program, and retrieve associated information corresponding to the user communication information from the insurance data system;

[0029] A processing module, used to evaluate the associated information based on a trained scoring model to obtain a recommendability score, wherein the scoring model is obtained by training a deep neural network based on the operation sample;

[0030] A generation module is used to determine a conversation keyword that matches the recommendable score and generate a recommendation conversation of product information based on the conversation keyword.

[0031] Furthermore, the device also includes:

[0032] A construction module is used to retrieve associated samples marked with different scores from the insurance data system and construct a multi-input and multi-output deep neural network, wherein the associated samples include transaction operation samples, identity samples, and channel operation samples;

[0033] A training module is used to perform model training on the deep neural network based on the operation sample to obtain the scoring model.

[0034] Furthermore, the training module is specifically used to adjust the loss function based on the communication value coefficient when performing model training on the deep neural network based on the operation sample, and calculate the scoring loss value based on the adjusted loss function, and the communication value coefficient is used to characterize the evaluation results of transaction behaviors associated with user communication information.

[0035] Furthermore, the generation module is specifically used to retrieve a conversation template, where the conversation template is obtained by combining transaction words and sentences in advance; the conversation keywords are inserted into the conversation template according to word order and word semantics to obtain a recommended conversation.

[0036] Furthermore, the generation module is specifically used to parse the recommendation strength score, transaction expectation score and conversation scenario score in the recommendable score; retrieve the preset score keyword mapping relationship, and query the conversation keywords that match the recommendable score based on the preset score keyword mapping relationship, wherein the preset score keyword mapping relationship includes conversation keywords corresponding to different recommendation strength scores, transaction expectation scores and conversation scenario scores, and the conversation keywords are used to characterize product keywords and recommendation speech keywords of insurance products.

[0037] Furthermore, the acquisition module is specifically used to send communication information acquisition instructions to each of the applications so that the application can feedback authorized user communication information; in response to the feedback instruction of the application, send the user communication information carried in the feedback instruction to the insurance data system so that the insurance data system queries the associated information matching the user communication information for feedback; and receive the associated information.

[0038] Furthermore, the device also includes:

[0039] A statistics module, used to count the product transaction results of the recommendation session;

[0040] A marking module is used to mark the user communication information based on the product transaction result to update the transaction operation information in the associated information.

[0041] According to another aspect of the present invention, a storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to perform operations corresponding to the above-mentioned conversation generation method for product information recommendation based on the scoring model.

[0042] According to another aspect of the present invention, there is provided a computer device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0043] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned conversation generation method for product information recommendation based on the scoring model.

[0044] By means of the above technical solution, the technical solution provided by the embodiment of the present invention has at least the following advantages:

[0045] The present invention provides a method and device for generating a conversation based on a scoring model for product information recommendation. Compared with the prior art, the embodiment of the present invention obtains user communication information entered by each application program, and retrieves associated information corresponding to the user communication information from the insurance data system; evaluates and processes the associated information based on a trained scoring model to obtain a recommendability score, wherein the scoring model is obtained by training a deep neural network based on operation samples; determines conversation keywords matching the recommendability score, and generates a recommended conversation of product information based on the conversation keywords, thereby achieving the purpose of flexible generation of the recommended conversation text, so that the product recommendation words are effectively updated based on user operations, more in line with the user's transaction needs, and avoids the singleness of the output of the product information recommendation conversation, thereby improving the effectiveness and accuracy of the intelligent conversation recommendation of insurance product information.

[0046] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0048] Figure 1 A flow chart of a method for generating a session for product information recommendation based on a scoring model provided by an embodiment of the present invention is shown;

[0049] Figure 2 A flowchart of another method for generating a session for product information recommendation based on a scoring model provided by an embodiment of the present invention is shown;

[0050] Figure 3 A block diagram showing the composition of a conversation generation device for product information recommendation based on a scoring model provided by an embodiment of the present invention;

[0051] Figure 4 A schematic structural diagram of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0052] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0053] When recommending insurance product information in a conversation, the fixed text sentences are single, which will lead to poor recommendation effect of insurance product information, and the selected insurance products cannot meet the flexible recommendation needs of different users, which greatly reduces the accuracy of recommending insurance product information through intelligent conversation. In the embodiment of the present invention, the subject of the conversation generation method for executing product information recommendation based on the scoring model can be a server. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0054] The embodiment of the present invention provides a method for generating a session for recommending product information based on a scoring model. Figure 1 As shown, the method includes:

[0055] 101. Obtain user communication information entered by each application program, and retrieve associated information corresponding to the user communication information from an insurance data system.

[0056] In the embodiment of the present invention, the server as the current execution end can be a session generation system, which interacts with the insurance data system and each application respectively. At this time, the user communication information is the communication information that represents the user identity, including but not limited to an email address or a telephone number, which is not specifically limited in the embodiment of the present invention. Among them, each application can send the user communication information to the current execution end after the user enters and authorizes to participate in the promotion of relevant information, and then retrieves the associated information matching the user communication information from the insurance data system of data interaction, wherein the associated information includes but is not limited to transaction operation information, identity information, and channel operation information. The transaction operation information is the operation information of the user completing similar or identical product transactions, such as the transaction time and number of transactions for purchasing insurance products, etc. The identity information is the user's real identity identification, such as an ID number, etc. The channel operation information is the information of the user logging in or registering different applications to operate the product information, such as the number of visits to view product information by the user logging in to platform a, etc., which is not specifically limited in the embodiment of the present invention.

[0057] It should be noted that for intelligent robots that are not clear about the expression content to follow up, the embodiments of the present invention can filter out the expression content that they want to continue to follow up based on the scoring of user communication information, and use it as a clue for intelligent conversation to make intelligent replies. Furthermore, the embodiments of the present invention can provide more accurate learning basis for intelligent conversation robots after scoring based on user communication information, better arrange time and resources, and help intelligent conversation robots quickly and accurately identify high-quality semantic expression content, thereby improving the conversion efficiency or closing rate of semantic expression content, accurately predicting semantic replies, and providing more accurate conversation reply information for product information recommendations.

[0058] 102. Evaluate the associated information based on the trained scoring model to obtain a recommendability score.

[0059] In the embodiment of the present invention, the scoring model is obtained by training a deep neural network based on the operation samples. The deep neural network is a multi-input and multi-output deep neural network (Deep Neural Network, DNN for short). The scoring model is obtained by pre-training the model, and the associated information is evaluated and processed to obtain a recommendability score. Among them, the recommendability score is used to characterize the evaluation possibility that the user communication information can be used as a conversation recommendation, and may include a recommendation strength score, a transaction expectation score, and a conversation scenario score, which is not specifically limited in the embodiment of the present invention. The deep neural network DNN includes an input layer (Input Layer) for receiving raw data, a hidden layer (Hidden Layers) for processing and feature extraction through multiple layers of neurons, and an output layer (Output Layer) for generating a final prediction result, wherein the hidden layer is usually composed of multiple layers of fully connected layers or convolutional layers. At the same time, in order to be able to learn complex patterns, an activation function Sigmoid can be selected, which is not specifically limited in the embodiment of the present invention.

[0060] It should be noted that when training a deep neural network DNN, the training set can be divided into observation values ​​and verification values, specifically including: 1. Data observation features with known label values ​​will be submitted to the input layer. At this time, the observations are grouped into multiple batches, such as usually divided into small batches; 2. Neurons work, that is, after passing through the input layer, the results are passed to the next layer after activation until the output layer generates a prediction; 3. The prediction is compared with the actual known value, and the difference between the predicted value and the true value (called "loss") is calculated; 4. Based on the results, revised values ​​of weights and bias values ​​are calculated to reduce losses, and these adjustments are back-propagated to neurons in the network layer; 5. In the next iteration round, batch training is repeated forward using the modified weights and bias values ​​to improve the accuracy of the model by reducing losses.

[0061] 103. Determine a conversation keyword that matches the recommendable score, and generate a recommendation conversation for product information based on the conversation keyword.

[0062] In an embodiment of the present invention, after the current execution end obtains a recommendable score, in order to improve the effectiveness of the recommendation session, the session keywords are determined based on the recommendable score matching. The session keywords are used to characterize product keywords and recommendation speech keywords of insurance products. Product keywords can be the names of different insurance products, product serial numbers, etc. Recommended speech keywords can be speech words for strong recommendations, general recommendations, and weaker recommendations. For example, strongly recommended speech words are "particularly suitable", "meet all needs", etc., generally recommended speech words are "good effect", "good benefits", etc., and weaker recommended speech words are "whether needed", "consider", etc., which are not specifically limited in the embodiment of the present invention.

[0063] It should be noted that, when determining the conversation keywords that match the recommendable scores, the conversation keywords that match different recommendable scores can be queried based on the pre-established mapping relationship. At this time, the pre-established mapping relationship can be configured by dividing into multiple score ranges, or one or more conversation keywords corresponding to each score can be configured to generate a recommendation conversation of product information based on the conversation keywords. Among them, the product information can include but is not limited to financial products, Internet products, industrial products, service products, etc., for example, commercial insurance, mechanical equipment, housekeeping services, etc., which are not specifically limited in the embodiments of the present invention.

[0064] The embodiment of the present invention provides a method for generating a conversation based on a scoring model for product information recommendation. Compared with the prior art, the embodiment of the present invention obtains user communication information entered by each application, and retrieves the associated information corresponding to the user communication information from the insurance data system; evaluates and processes the associated information based on the trained scoring model to obtain a recommendability score, and the scoring model is obtained by training a deep neural network based on operation samples; determines the conversation keywords that match the recommendability score, and generates a recommended conversation of product information based on the conversation keywords, so as to achieve the purpose of flexible generation of the recommended conversation text, so that the product recommendation words are effectively updated based on user operations, more in line with the user's transaction needs, and avoid the singleness of the output of the product information recommendation conversation, thereby improving the effectiveness and accuracy of the intelligent conversation recommendation of insurance product information.

[0065] The embodiment of the present invention provides a method for generating a session for recommending product information based on a scoring model. Figure 2 As shown, the method includes:

[0066] 201. Obtain user communication information entered by each application program, and retrieve associated information corresponding to the user communication information from an insurance data system.

[0067] Among them, the associated information includes but is not limited to transaction operation information, identity information, and channel operation information. The transaction operation information is the operation information of the user completing similar or identical product transactions, such as the transaction time and number of transactions for purchasing insurance products. The identity information is the user's real identity identification, such as an ID number. The channel operation information is the information of the user logging in or registering different applications to operate on product information, such as the number of visits by the user to platform a to view product information, etc. The embodiment of the present invention does not make specific limitations.

[0068] In another embodiment of the present invention, for further definition and explanation, step 201 obtains user communication information entered by each application, and retrieves associated information corresponding to the user communication information from the insurance data system, including:

[0069] Sending a communication information acquisition instruction to each of the applications, so that the application can feedback authorized user communication information;

[0070] In response to the feedback instruction of the application, sending the user communication information carried in the feedback instruction to the insurance data system, so that the insurance data system queries the associated information matching the user communication information for feedback;

[0071] The association information is received.

[0072] Since the current execution end is only the subject of the session generation execution, in order to ensure the security of insurance data and user information, when the current execution end obtains the associated information corresponding to the user communication information, it first sends a communication information acquisition instruction to the application so that the application can feedback the authorized user communication information. Among them, the application can be a social platform or an electronic commodity trading platform, etc. The user logs in to the application to browse videos, place orders for commodities, etc., retains the user communication information, and after choosing to authorize the use of the user communication information, the user communication information can be fed back to the current execution end based on the communication information acquisition instruction, such as a telephone number or an email address, etc., which is not specifically limited in the embodiment of the present invention. After receiving the feedback user communication information, the current execution end sends this user communication information to the insurance data system so that the insurance data system queries the associated information matching this user communication information and feeds it back to the current end. In addition, if the associated information matching the user communication information cannot be found, the associated information can be configured as empty or a value of 0, so that the current execution end can directly evaluate based on this information.

[0073] 202. Retrieve relevant samples with different scores from the insurance data system and build a multi-input and multi-output deep neural network.

[0074] In order to evaluate the associated information based on intelligent learning technology, so as to serve as the basis for generating recommended conversations, the current execution end first retrieves associated samples marked with different scores from the insurance data system before performing the evaluation process. At this time, the associated samples include transaction operation samples, identity samples, and channel operation samples. Among them, the transaction operation samples are operation samples of users completing similar or identical product transactions, the identity samples are samples of users' real identity identification, and the channel operation samples are samples of users logging in or registering different applications to operate on product information. The embodiments of the present invention do not make specific limitations. At the same time, a deep neural network with multiple inputs and multiple outputs is constructed. As a neural network containing multiple hidden layers, the deep neural network can learn more complex features and patterns. At this time, the deep neural network DNN includes an input layer (Input Layer) for receiving raw data, a hidden layer (Hidden Layers) for processing and feature extraction through multiple layers of neurons, and an output layer (Output Layer) for generating the final prediction result.

[0075] 203. Perform model training on the deep neural network based on the operation sample to obtain the scoring model.

[0076] In order to accurately train the scoring model to improve the accuracy of the scoring of user communication information, the hidden layer is usually composed of multiple layers of fully connected layers or convolutional layers. At the same time, in order to be able to learn complex patterns, the activation function Sigmoid can be selected, which is not specifically limited in the embodiment of the present invention. In a specific implementation scenario, the deep neural network in the embodiment of the present invention is further defined as a feedforward neural network, that is, information is transmitted from the input layer to the output layer through a series of hidden layers, and each layer of neurons is connected to all neurons in the next layer to form a fully connected layer for regression and classification tasks, and the loss function is selected as the cross-entropy loss (Cross-Entroyp Loss) to perform classification evaluation tasks. Furthermore, after the deep neural network is constructed, the deep neural network is model trained based on the operation sample to obtain a scoring model. Specifically, the operation sample needs to be preprocessed, including but not limited to preprocessing, data enhancement, data cleaning, etc., which is not specifically limited in the embodiment of the present invention. In addition, a multi-output multi-input deep neural network is used in the embodiment of the present invention. During learning, the training samples are marked with training marks for recommendation strength score, transaction expectation score and conversation scenario score, so that the scoring model that has completed model training outputs three scoring results. At this time, the recommendation strength score, transaction expectation score and conversation scenario score can be marked manually during model training, and the embodiment of the present invention does not make specific limitations.

[0077] In another embodiment of the present invention, for further definition and explanation, the step of performing model training on the deep neural network based on the operation sample to obtain the scoring model includes:

[0078] When the deep neural network is trained based on the operation sample, the loss function is adjusted based on the communication value coefficient, and the scoring loss value is calculated based on the adjusted loss function.

[0079] In order to optimize the scoring model, make the model more adaptable, and improve the learning effect of the model, the current execution end adjusts the loss function based on the communication value coefficient when training the model. At this time, the communication value coefficient is used to characterize the evaluation results of the transaction behavior associated with the user communication information. At this time, the transaction behavior is the behavior of trading insurance products, and the transaction behavior associated with the user communication information is the behavior of binding the transaction insurance product according to the user communication information. For example, the transaction behavior of placing an order to purchase an insurance product by using a telephone number as the order number is not specifically limited in the embodiment of the present invention. At this time, the communication value coefficient as the evaluation result can be pre-marked with an evaluation score for the transaction behavior. In the range of 0-1, the closer to 1, the higher the evaluation score. It can be configured based on the payment speed and payment amount. The embodiment of the present invention does not make specific limitations.

[0080] 204. Evaluate the associated information based on the trained scoring model to obtain a recommendability score.

[0081] In an embodiment of the present invention, when determining a conversation keyword that matches a recommendable score, the conversation keywords that match different recommendable scores can be queried based on a pre-established mapping relationship. At this time, the pre-established mapping relationship can be configured by dividing into multiple score ranges, or one or more conversation keywords corresponding to each score can be configured to generate a recommendation conversation of product information based on the conversation keywords. Among them, the product information can include but is not limited to financial products, Internet products, industrial products, service products, etc., for example, commercial insurance, mechanical equipment, housekeeping services, etc., which are not specifically limited in the embodiment of the present invention. And step 203 is the same as step 102, and will not be repeated here.

[0082] 205. Analyze the recommendation strength score, transaction expectation score, and conversation scenario score in the recommendable score.

[0083] In order to achieve more efficient conversation keyword screening, the current execution end first analyzes the recommendation strength score, transaction expectation score and conversation scenario score in the recommendable score. Among them, the recommendation strength score is used to represent the score of recommendation enthusiasm, the transaction expectation score is used to represent the score of transaction possibility, and the conversation scenario score is used to represent the score of the effect of participating in the conversation recommendation. The current execution end can call the preset score keyword mapping relationship and query the conversation keyword matching the recommendable score based on this preset score keyword mapping relationship. The preset score keyword mapping relationship includes conversation keywords corresponding to different recommendation strength scores, transaction expectation scores and conversation scenario scores.

[0084] 206 . Retrieve a preset scoring keyword mapping relationship, and query a conversation keyword matching the recommendable score based on the preset scoring keyword mapping relationship.

[0085] It should be noted that the preset scoring keyword mapping relationship includes conversation keywords corresponding to different recommendation strength scores, transaction expectation scores, and conversation scenario scores, and the conversation keywords are used to characterize product keywords and recommendation speech keywords of insurance products. Since conversation keywords are used to characterize product keywords and recommendation speech keywords of insurance products, when matching conversation keywords based on the recommendability score, the preset scoring keyword mapping relationship can be configured with the latest product keywords and the strongest recommendation speech words for scores with higher scores, which are not specifically limited in the embodiments of the present invention.

[0086] 207. Retrieve the session template.

[0087] In order to meet the flexible generation of different conversations, thereby speeding up the efficiency of conversation generation, the current execution end pre-calls the conversation template, and at this time, the conversation template is obtained by pre-combining transaction language sentences. Among them, the transaction language sentences can be obtained by combining multiple words obtained by manually splitting common transaction language sentences. Different common transaction language sentences can be pre-constructed for different product information, and the embodiment of the present invention does not make specific limitations.

[0088] 208. Insert the conversation keywords into the conversation template according to word order and word semantics to obtain a recommended conversation.

[0089] In order to achieve the retrospective nature of the conversation recommendation, after the current execution end generates the recommendation conversation, the current execution end sends the recommendation conversation to the recommendation system for execution. At this time, the recommendation can be made by dialing based on a voice robot or by sending based on an email, which is not specifically limited in the embodiment of the present invention. In addition, the transaction results of the user's product transaction based on the recommendation conversation are obtained, for example, the purchase of an insurance product or the expected purchase without purchase, etc., and then, the user communication information is marked based on the product transaction result, and the transaction operation information corresponding to the user communication information is updated, so that the content of the recommendation conversation can be changed when the conversation is generated next time, thereby forming a virtuous cycle of conversation recommendation.

[0090] In another embodiment of the present invention, for further definition and explanation, before the step of evaluating the associated information based on the trained scoring model to obtain the recommendability score, the method further includes:

[0091] Retrieve relevant samples with different scores from the insurance data system and build a multi-input and multi-output deep neural network;

[0092] The deep neural network is trained based on the operation samples to obtain the scoring model.

[0093] In order to evaluate the associated information based on intelligent learning technology, so as to serve as the basis for generating recommended conversations, the current execution end first retrieves associated samples marked with different scores from the insurance data system before performing the evaluation process. At this time, the associated samples include transaction operation samples, identity samples, and channel operation samples. Among them, the transaction operation samples are operation samples of users completing similar or identical product transactions, the identity samples are samples of users' real identity identification, and the channel operation samples are samples of users logging in or registering different applications to operate product information. The embodiment of the present invention does not make specific restrictions. At the same time, a multi-input and multi-output deep neural network is constructed. As a neural network containing multiple hidden layers, the deep neural network can learn more complex features and patterns. At this time, the deep neural network DNN includes an input layer (Input Layer) for receiving raw data, a hidden layer (Hidden Layers) for processing and feature extraction through multiple layers of neurons, and an output layer (Output Layer) for generating the final prediction result. Among them, the hidden layer is usually composed of multiple layers of fully connected layers or convolutional layers. At the same time, in order to be able to learn complex patterns, the activation function Sigmoid can be selected. The embodiment of the present invention does not make specific restrictions. In a specific implementation scenario, the deep neural network in the embodiment of the present invention is further defined as a feedforward neural network, that is, information is transmitted from the input layer to the output layer through a series of hidden layers, and each layer of neurons is connected to all neurons in the next layer to form a fully connected layer for regression and classification tasks, and the loss function is selected as the cross-entropy loss (Cross-Entroyp Loss) to perform classification evaluation tasks. Furthermore, after the deep neural network is constructed, the deep neural network is trained based on the operation sample to obtain a scoring model. Specifically, the operation sample needs to be preprocessed, including but not limited to preprocessing, data enhancement, data cleaning, etc., which are not specifically limited in the embodiment of the present invention. In addition, a multi-output multi-input deep neural network is used in the embodiment of the present invention. When learning, the training samples are marked with training marks for recommendation strength scoring, transaction expectation scoring, and conversation scene scoring, so that the scoring model that completes the model training outputs three scoring results. At this time, the recommendation strength scoring, transaction expectation scoring, and conversation scene scoring can be marked based on manual methods when the model is trained, which are not specifically limited in the embodiment of the present invention.

[0094] In another embodiment of the present invention, for further definition and explanation, after the step of generating a recommendation session of product information based on the session keywords, the method further includes:

[0095] Counting the product transaction results of the recommendation session;

[0096] The user communication information is marked based on the product transaction result to update the transaction operation information in the associated information.

[0097] In order to achieve the retrospective nature of the conversation recommendation, after the current execution end generates the recommendation conversation, the current execution end sends the recommendation conversation to the recommendation system for execution. At this time, the recommendation can be made by dialing based on a voice robot or by sending based on an email, which is not specifically limited in the embodiment of the present invention. In addition, the transaction results of the user's product transaction based on the recommendation conversation are obtained, for example, the purchase of an insurance product or the expected purchase without purchase, etc., and then, the user communication information is marked based on the product transaction result, and the transaction operation information corresponding to the user communication information is updated, so that the content of the recommendation conversation can be changed when the conversation is generated next time, thereby forming a virtuous cycle of conversation recommendation.

[0098] The embodiment of the present invention provides a method for generating a conversation based on a scoring model for product information recommendation. Compared with the prior art, the embodiment of the present invention obtains user communication information entered by each application, and retrieves the associated information corresponding to the user communication information from the insurance data system; evaluates and processes the associated information based on the trained scoring model to obtain a recommendability score, and the scoring model is obtained by training a deep neural network based on operation samples; determines the conversation keywords that match the recommendability score, and generates a recommended conversation of product information based on the conversation keywords, so as to achieve the purpose of flexible generation of the recommended conversation text, so that the product recommendation words are effectively updated based on user operations, more in line with the user's transaction needs, and avoid the singleness of the output of the product information recommendation conversation, thereby improving the effectiveness and accuracy of the intelligent conversation recommendation of insurance product information.

[0099] Furthermore, as a response to the above Figure 1 The embodiment of the present invention provides a conversation generation device for product information recommendation based on a scoring model, such as Figure 3 As shown, the device comprises:

[0100] An acquisition module 31 is used to acquire user communication information entered by each application program, and retrieve associated information corresponding to the user communication information from the insurance data system;

[0101] A processing module 32 is used to evaluate the associated information based on a trained scoring model to obtain a recommendability score, wherein the scoring model is obtained by training a deep neural network based on the operation sample;

[0102] The generating module 33 is used to determine a conversation keyword matching the recommendable score, and generate a recommendation conversation of the product information based on the conversation keyword.

[0103] Furthermore, the device also includes:

[0104] A construction module is used to retrieve associated samples marked with different scores from the insurance data system and construct a multi-input and multi-output deep neural network, wherein the associated samples include transaction operation samples, identity samples, and channel operation samples;

[0105] A training module is used to perform model training on the deep neural network based on the operation sample to obtain the scoring model.

[0106] Furthermore, the training module is specifically used to adjust the loss function based on the communication value coefficient when performing model training on the deep neural network based on the operation sample, and calculate the scoring loss value based on the adjusted loss function, and the communication value coefficient is used to characterize the evaluation results of transaction behaviors associated with user communication information.

[0107] Furthermore, the generation module is specifically used to retrieve a conversation template, where the conversation template is obtained by combining transaction words and sentences in advance; the conversation keywords are inserted into the conversation template according to word order and word semantics to obtain a recommended conversation.

[0108] Furthermore, the generation module is specifically used to parse the recommendation strength score, transaction expectation score and conversation scenario score in the recommendable score; retrieve the preset score keyword mapping relationship, and query the conversation keywords that match the recommendable score based on the preset score keyword mapping relationship, wherein the preset score keyword mapping relationship includes conversation keywords corresponding to different recommendation strength scores, transaction expectation scores and conversation scenario scores, and the conversation keywords are used to characterize product keywords and recommendation speech keywords of insurance products.

[0109] Furthermore, the acquisition module is specifically used to send communication information acquisition instructions to each of the applications so that the application can feedback authorized user communication information; in response to the feedback instruction of the application, send the user communication information carried in the feedback instruction to the insurance data system so that the insurance data system queries the associated information matching the user communication information for feedback; and receive the associated information.

[0110] Furthermore, the device also includes:

[0111] A statistics module, used to count the product transaction results of the recommendation session;

[0112] A marking module is used to mark the user communication information based on the product transaction result to update the transaction operation information in the associated information.

[0113] The embodiment of the present invention provides a device for generating a conversation based on a scoring model for product information recommendation. Compared with the prior art, the embodiment of the present invention obtains user communication information entered by each application, and retrieves the associated information corresponding to the user communication information from the insurance data system; evaluates and processes the associated information based on the trained scoring model to obtain a recommendability score, wherein the scoring model is obtained by training a deep neural network based on operation samples; determines conversation keywords matching the recommendability score, and generates a recommended conversation of product information based on the conversation keywords, so as to achieve the purpose of flexible generation of the recommended conversation text, so that the product recommendation words are effectively updated based on user operations, more in line with the user's transaction needs, and avoid the singleness of the output of the product information recommendation conversation, thereby improving the effectiveness and accuracy of the intelligent conversation recommendation of insurance product information.

[0114] According to an embodiment of the present invention, a storage medium is provided, wherein the storage medium stores at least one executable instruction, and the computer executable instruction can execute the session generation method for product information recommendation based on a scoring model in any of the above method embodiments.

[0115] Figure 4 A schematic diagram of the structure of a computer device provided according to an embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computer device.

[0116] like Figure 4 As shown, the computer device may include: a processor 402 , a communication interface 404 , a memory 406 , and a communication bus 408 .

[0117] The processor 402 , the communication interface 404 , and the memory 306 communicate with each other via a communication bus 408 .

[0118] The communication interface 404 is used to communicate with other devices such as clients or other servers.

[0119] The processor 402 is used to execute the program 410, and specifically can execute the relevant steps in the embodiment of the conversation generation method for product information recommendation based on the above-mentioned scoring model.

[0120] Specifically, the program 410 may include program codes, which include computer operation instructions.

[0121] The processor 402 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiment of the present invention. The one or more processors included in the computer device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0122] The memory 406 is used to store the program 410. The memory 406 may include a high-speed RAM memory, and may also include a non-volatile memory, such as at least one disk memory.

[0123] The program 410 may be specifically configured to enable the processor 402 to perform the following operations:

[0124] Obtaining user communication information entered by each application, and retrieving associated information corresponding to the user communication information from the insurance data system;

[0125] The associated information is evaluated based on a trained scoring model to obtain a recommendability score, wherein the scoring model is obtained by training a deep neural network based on the operation sample;

[0126] A conversation keyword matching the recommendable score is determined, and a recommendation conversation of product information is generated based on the conversation keyword.

[0127] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0128] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for generating a conversation for product information recommendation based on a scoring model, characterized in that: include: Obtaining user communication information entered by each application, and retrieving associated information corresponding to the user communication information from the insurance data system; The associated information is evaluated based on a trained scoring model to obtain a recommendability score, wherein the scoring model is obtained by training a deep neural network based on the operation sample; A conversation keyword matching the recommendable score is determined, and a recommendation conversation of product information is generated based on the conversation keyword.

2. The method according to claim 1, characterized in that Before evaluating the associated information based on the trained scoring model to obtain a recommendability score, the method further includes: Retrieving associated samples marked with different scores from the insurance data system and constructing a multi-input and multi-output deep neural network, wherein the associated samples include transaction operation samples, identity samples, and channel operation samples; The deep neural network is trained based on the operation samples to obtain the scoring model.

3. The method according to claim 2, characterized in that The performing model training on the deep neural network based on the operation sample to obtain the scoring model comprises: When the deep neural network is trained based on the operation sample, the loss function is adjusted based on the communication value coefficient, and the scoring loss value is calculated based on the adjusted loss function. The communication value coefficient is used to characterize the evaluation results of transaction behaviors associated with user communication information.

4. The method according to claim 1, characterized in that: The recommendation session for generating product information based on the session keywords includes: Retrieving a conversation template, wherein the conversation template is obtained by combining transaction speech sentences in advance; The conversation keywords are inserted into the conversation template according to the word order and word semantics to obtain a recommended conversation.

5. The method according to claim 4, characterized in that The determining of the conversation keywords matching the recommendable score comprises: Analyze the recommendation strength score, transaction expectation score, and conversation scenario score in the recommendable score; Retrieve a preset scoring keyword mapping relationship, and query the conversation keywords that match the recommendable score based on the preset scoring keyword mapping relationship. The preset scoring keyword mapping relationship includes conversation keywords corresponding to different recommendation strength scores, transaction expectation scores, and conversation scenario scores. The conversation keywords are used to characterize product keywords and recommendation speech keywords of insurance products.

6. The method according to claim 1, characterized in that The obtaining of user communication information entered by each application program and retrieving associated information corresponding to the user communication information from the insurance data system includes: Sending a communication information acquisition instruction to each of the applications, so that the application can feedback authorized user communication information; In response to the feedback instruction of the application, sending the user communication information carried in the feedback instruction to the insurance data system, so that the insurance data system queries the associated information matching the user communication information for feedback; The association information is received.

7. The method according to claim 1, characterized in that After generating a recommendation session of product information based on the session keywords, the method further includes: Counting the product transaction results of the recommendation session; The user communication information is marked based on the product transaction result to update the transaction operation information in the associated information.

8. A conversation generation device for product information recommendation based on a scoring model, characterized in that: include: An acquisition module, used to acquire user communication information entered by each application program, and retrieve associated information corresponding to the user communication information from the insurance data system; A processing module, used to evaluate the associated information based on a trained scoring model to obtain a recommendability score, wherein the scoring model is obtained by training a deep neural network based on the operation sample; A generation module is used to determine a conversation keyword that matches the recommendable score and generate a recommendation conversation of product information based on the conversation keyword.

9. A storage medium storing at least one executable instruction, wherein the executable instruction enables a processor to execute operations corresponding to the conversation generation method for product information recommendation based on a scoring model as described in any one of claims 1 to 7.

10. A computer device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the conversation generation method for product information recommendation based on a scoring model as described in any one of claims 1 to 7.

Images