CN113034295A

CN113034295A - Dangerous species recommendation method and device, electronic equipment and storage medium

Info

Publication number: CN113034295A
Application number: CN202110428931.1A
Authority: CN
Inventors: 许国良; 陈青山; 郑晓华; 康祖荫
Original assignee: China Construction Bank Corp
Current assignee: China Construction Bank Corp
Priority date: 2021-04-21
Filing date: 2021-04-21
Publication date: 2021-06-25

Abstract

The embodiment of the invention relates to the technical field of artificial intelligence, and particularly relates to and discloses a dangerous seed recommendation method and device, electronic equipment and a storage medium. The method comprises the following steps: determining target users for recommending at least two target long-term risk types, respectively predicting purchasing tendency of the target long-term risk types through an application party federal half model generated by a data application party and a provider federal half model generated by a data provider, and determining a target insurance application tendency prediction result of the target users on the target long-term risk types; and recommending target long-term dangerous seed products to the target user according to the target insurance tendency prediction result of the target user on each target long-term dangerous seed. By adopting the scheme, the data application party and the data provider only exchange intermediate results but not exchange original data through federal learning, so that invisible model training for data can be achieved, data combined modeling is achieved on the premise of protecting data privacy, prediction results are aggregated by the prediction results, and recommendation of dangerous products is achieved.

Description

Dangerous species recommendation method and device, electronic equipment and storage medium

Technical Field

The embodiment of the invention relates to the technical field of artificial intelligence, in particular to a dangerous seed recommendation method and device, electronic equipment and a storage medium.

Background

With the continuous development of cloud computing technology and big data technology, organizations and individuals can continuously generate a large amount of data. Particularly in the accurate acquisition of customer and insurance pricing links, multi-dimensional data support is usually required, such as including qualification information, purchasing ability, physical conditions and the like of consumers, so as to provide customized insurance products and services for different consumer groups.

However, since data contains a lot of personal privacy, business confidentiality and the like, different organizations or individuals usually do not provide data grasped by themselves to the outside, such "data barriers" form a lot of "data islands", which results in combination and imperfection of big data and artificial intelligence, and data value is not sufficiently mined and released, thereby having a great influence on accurate acquisition of customers and insurance pricing.

Disclosure of Invention

The embodiment of the invention provides a dangerous case recommendation method, a dangerous case recommendation device, electronic equipment and a storage medium, and aims to accurately acquire customer and insurance pricing.

In a first aspect, an embodiment of the present invention provides a dangerous seed recommendation method, which is executed by a data application side, and the method includes:

determining a target user for carrying out at least two kinds of target long-term risk recommendation;

respectively predicting the purchasing tendency of the target long-term risk through an application party federal half model generated by a data application party and a provider federal half model generated by a data provider, and determining the target insurance tendency prediction result of the target user on the target long-term risk;

and recommending target long-term dangerous seed products to the target user according to the target insurance tendency prediction result of the target user on each target long-term dangerous seed.

In a second aspect, an embodiment of the present invention further provides an dangerous seed recommendation apparatus configured on a data application side, where the apparatus includes:

the target user determination module is used for determining target users for recommending at least two target long-term dangerous species;

the purchasing tendency prediction module is used for respectively predicting purchasing tendency of the target long-term risk variety through an application party federal half model generated by a data application party and a provider federal half model generated by a data provider, and determining a target insurance application tendency prediction result of the target user on the target long-term risk variety;

and the dangerous seed product recommending module is used for recommending the target long-term dangerous seed products to the target user according to the target insurance tendency forecasting result of the target user on each target long-term dangerous seed.

In a third aspect, an embodiment of the present invention further provides an electronic device, including:

one or more processors;

storage means for storing one or more programs;

the one or more programs are executable by the one or more processors to cause the one or more processors to implement a seed risk recommendation method as provided in any embodiment of the invention.

In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the seed risk recommendation method as provided in any of the embodiments of the present invention.

The embodiment of the invention provides a dangerous seed recommendation method, which comprises the steps of determining a target user for recommending at least two target long-term dangerous seeds, respectively predicting the purchasing tendency of the target long-term dangerous seeds through an application party federal half model generated by a data application party and a provider federal half model generated by a data provider, and determining a target application tendency prediction result of the target user on the target long-term dangerous seeds; and recommending target long-term dangerous seed products to the target user according to the target insurance tendency prediction result of the target user on each target long-term dangerous seed. By adopting the scheme, the data application party and the data provider only exchange intermediate results but not original data through a federal learning algorithm, so that the data can be invisible to carry out respective model training, and the data combined modeling is realized on the premise of protecting the data privacy; and the data application party and the data provider respectively use the trained half models to predict based on own data, and then collect the respective prediction results to one party to aggregate the prediction results, so that the recommendation of the dangerous seed products is realized.

The above summary of the present invention is merely an overview of the technical solutions of the present invention, and the present invention can be implemented in accordance with the content of the description in order to make the technical means of the present invention more clearly understood, and the above and other objects, features, and advantages of the present invention will be more clearly understood.

Drawings

Other features, objects and advantages of the invention will become more apparent upon reading of the following detailed description of non-limiting embodiments thereof, made with reference to the accompanying drawings. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:

FIG. 1 is a flow chart of a risk recommendation method provided in an embodiment of the present invention;

FIG. 2 is a schematic diagram of federal learning in an emergency recommendation scenario provided in an embodiment of the present invention;

FIG. 3 is a schematic process diagram of modeling of a data application side in a longitudinal federated learning process provided in an embodiment of the present invention;

FIG. 4 is a schematic process diagram of modeling of a data provider side in a longitudinal federated learning process provided in an embodiment of the present invention;

FIG. 5 is a schematic diagram of a positive sample and a negative sample obtained according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of a process for forecasting an application tendency provided in an embodiment of the present invention;

FIG. 7 is a block diagram of a dangerous seed recommending apparatus provided in an embodiment of the present invention;

fig. 8 is a schematic structural diagram of an electronic device provided in an embodiment of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the accompanying drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be further noted that, for the convenience of description, only some of the structures related to the present invention are shown in the drawings, not all of the structures.

Before discussing exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although a flowchart may describe the operations (or steps) as a sequential process, many of the operations can be performed in parallel, concurrently or simultaneously. In addition, the order of the operations may be re-arranged. The process may be terminated when its operations are completed, but may have additional steps not included in the figure. The processes may correspond to methods, functions, procedures, subroutines, and the like.

For better understanding of the present application scheme, the following is a detailed explanation of the importance of data sharing support in some business scenarios. Integration, cross-industry and cross-field data sharing and application become the need, and more business scenes need multi-party data sharing to release the application value of data. Taking the financial industry as an example, a plurality of business scenes need to be combined with the joint modeling of external data, including digital marketing, wind control and anti-fraud, customer activation and credit analysis and the like; in addition, taking the insurance industry as an example, in the links of accurate customer acquisition and insurance pricing, multi-dimensional data support including qualification information, purchasing ability, physical conditions and the like of consumers is needed so as to provide customized insurance products and services for different consumer groups. In the multi-dimensional data supporting process, a multi-party secure data sharing system needs to be established, so that the data security problem occurring in the dynamic data using and sharing process can be avoided.

Hereinafter, a method, an apparatus, an electronic device, and a storage medium for recommending a seed risk provided in the embodiments of the present invention are described in detail through the embodiments and alternative solutions of the embodiments.

Fig. 1 is a flowchart of an emergency recommendation method provided in an embodiment of the present invention, and the technical solution of this embodiment may be applied to a case of recommending a suitable emergency product to a suitable user. The method can be executed by the risk category recommendation device, and the device can be realized by software and/or hardware and is integrated on any electronic equipment with network communication function. As shown in fig. 1, the dangerous seed recommendation method in the embodiment of the present invention may include the following steps:

and S110, determining target users for carrying out at least two kinds of target long-term risk recommendation.

In an insurance industry scenario, a short term risk category may refer to an insurance category with a short insurance limit, such as an insurance category with an insurance limit of one month or less (including one year); the long-term risk category may refer to an insurance category having a longer insurance period than that of the short-term risk category, for example, an insurance category having an insurance period of one year or more (one year or less). Because the risk category recommendation is not necessarily made for all long-term risk categories, the target long-term risk category can be a risk category which needs to be targeted for insurance product recommendation and is selected from a plurality of long-term risk categories when an insurance marketing strategy needs to be formulated.

And S120, respectively predicting the purchasing tendency of the target long-term risk through an application party federal half model generated by a data application party and a provider federal half model generated by a data provider, and determining the target insurance tendency prediction result of the target user on the target long-term risk.

The data application party can know customers who buy short-term dangerous species through a direct marketing channel and can also know customers who only buy a single dangerous species product partially, and the data application party hopes to recommend suitable/buying-prone dangerous species products for the customers through the purchasing behavior characteristics, insurance product characteristics and user attribute characteristics of insurance products by the users, so that the success rate of secondary marketing is improved. However, in order to ensure data security, data of a data provider cannot be directly provided to a data application party, which results in that the data application party cannot utilize the user attribute characteristics to realize recommendation of appropriate dangerous goods to a client.

Therefore, referring to fig. 2, the data application party can utilize the locally stored user to perform longitudinal federal learning on the insurance product purchasing behavior feature and insurance product feature combined data provider to obtain a federal half model corresponding to the data application party; meanwhile, the data provider can utilize the locally stored user attribute characteristics to perform longitudinal federal learning in combination with the data application party, so that a federal half model corresponding to the data provider is obtained.

By adopting the mode, under the condition that data of a data provider or a data application party is not changed, model training is respectively carried out on the data provider and the data application party by only exchanging intermediate results but not exchanging original data through a federal learning algorithm, and the richness of data characteristics in the model training process is improved, so that 'data is not available and not visible', the data privacy is guaranteed, and the model effect is improved by using multi-party data.

Referring to fig. 2, as an alternative, the application part federal semi-model generated by the data application part may be obtained by longitudinally federally learning, by the data application part, the user attribute characteristics under the first time window provided by the joint data provider based on the characteristics of the insurance product purchasing behavior and the characteristics of the insurance product of the user under the first time window provided locally under the federal learning framework. Correspondingly, the provider federal semi-model generated by the provider is obtained by longitudinally and federally learning the insurance product purchasing behavior characteristics and the insurance product characteristics by the user under the first time window provided by the combined data application party on the basis of the user attribute characteristics under the first time window provided by the data provider under the federal learning framework.

According to the scheme, data of a data application side is independently used for modeling, data of the data application side and data of a data provider side are used for centralized modeling (the data are placed in the same environment and are mutually visible), and data of the data application side and the data are used for federal modeling. According to experimental results, compared with single-party modeling and multi-party modeling, the characteristics are richer, and the effect of the model can be improved; compared with single-party modeling and federal modeling, one party has better characteristics when the federal modeling is carried out, and the effect of the model can be obviously improved; and comparing multi-party and federal modeling, federal modeling can cause some loss of accuracy. Thus, the federal study can be effectively applied to cooperate with the private data. Therefore, the precision loss of the federal dangerous recommended tasks is limited, the data privacy can be guaranteed, and the modeling effect can be achieved.

Optionally, since the time limit of the user for the insurance product purchasing behavior feature, the insurance product feature and the user attribute feature may change with time, the time limit of the user for the insurance product purchasing behavior feature, the insurance product feature and the user attribute feature used for model training may be defined through the first time window, and it is avoided that the trained model loses timeliness due to the use of the model training feature with lower timeliness, thereby affecting the prediction accuracy of the model. The user attribute characteristics stored by the data provider can include user basic information, user consumption messages, user asset information, user behavior information and the like.

Referring to fig. 3, as an alternative, a configuration uses the Secureboost federal safety elevated tree algorithm for model training on the data application side, and predicts the purchasing tendency as a classification problem. And the data application party establishes an emergency recommending task, configures the federal learning environment and informs the data provider of the IP and the ID of the federal learning environment. And processing the training sample set and the evaluation sample set into files by the data application party, wherein each sample record comprises a record ID + a user ID + an adventure feature, and the record ID is determined by an MD5 value obtained by hashing the user ID and the adventure ID. The data application party can develop the federal model recommended by the dangerous seeds and carry out model training on the basis of a local training sample set and an evaluation sample set of the data application party to generate the federal half model.

Referring to fig. 3, after model training is finished, the risk prediction is performed by using the locally generated federal semi-model to obtain a prediction result, and the prediction result comprises a risk ID and a risk prediction label. After obtaining the prediction result, the prediction score and the user ID may be associated with the risk ID according to the risk ID, so as to obtain the prediction result of the purchasing tendency of the user to the risk, such as the purchasing tendency probability of each user to all the risks, and set a corresponding rule to recommend the risk to each user.

Referring to fig. 4, as an alternative, model training at the data provider using the Secureboost federal safety elevated tree algorithm can also be configured to predict buying tendency as a classification problem. Establishing a dangerous seed recommendation task by a data provider; and configuring the federal learning environment, and informing the data application side of the federal learning environment IP and the ID. The data provider combines the dangerous seed ID provided by the data application party and the user ID of the data provider to generate a record ID, and the corresponding user ID forms a record in each record ID and falls into a file; uploading the file to a federated learning framework of the data provider. The data provider can develop a federal model recommended by the dangerous seeds, and model training is carried out on the basis of a training sample set and an evaluation sample set local to the data provider to generate a federal half model. After model training is finished, carrying out risk prediction by using a locally generated federal semi-model to obtain a prediction result, wherein the prediction result comprises a risk ID and a risk prediction label.

By adopting the mode, the two parties of the data application party and the data providing party prepare the local data of the two parties by virtue of a federal learning framework, and the federal learning modeling is carried out through the steps of data uploading, feature processing, model development, model training, model batch prediction, model real-time prediction and the like. The data application side uses the user characteristic data of the data provider to improve the richness of the data characteristics, only intermediate results are exchanged without original data through a federal learning algorithm, and therefore the data can be invisible, namely the data privacy is guaranteed, and the model effect can be improved by using multi-party data.

Referring to FIG. 5, as an alternative, for the data applicator and data provider, the training sample set and the evaluation sample set may include positive samples and negative samples, for example, for positive samples, an observation period user may be taken, (U, P, B) an insurance product P is purchased by a user U in a performance period, and a value of 1 for B represents a purchase; for the negative example, take the observation period user, (U, P, B) user U did not purchase insurance product P during the presentation period, where P excludes other insurance products within the broad category of insurance products to which P belongs.

As an alternative, the training samples and the evaluation samples may be processed as follows: confirming the sample data missing rate, judging the characteristic types (including a discrete type and a continuous type), filling the sample data missing (filling the discrete type by using-1 and filling the continuous type by using a median), encoding WOE, screening an IV value, checking the logicality and screening again according to the service of the model. Optionally, by performing business screening according to business experience, the data application party selects 59 discrete features and 5 continuous features for 64 features in total, and the data provider party selects 29 discrete features and 90 continuous features for 119 features in total. After further feature engineering, the data application side retains 28 discrete features and 5 continuous features related to the business, and the data provider side retains 6 discrete features and 88 continuous features for completing subsequent model construction. For example, sample data obtained in different time periods across the time periods are taken to respectively form a training sample set and an evaluation sample set, and the proportion of positive samples to negative samples in the training sample set and the evaluation sample set is ensured to be 1: 1.

As an alternative, the application party federal semi-model and the provider party federal semi-model respectively output the prediction results of the purchasing tendency of the target long-term risk under the corresponding model input characteristic dimension. For example, the data application party performs model input on the dimensions of the purchasing behavior characteristics and the characteristics of the insurance products of the users, and then the federal semi-model of the application party outputs the purchasing tendency prediction results of the target long-term risk types under the dimensions of the purchasing behavior characteristics and the characteristics of the insurance products of the corresponding users; and the data provider inputs a model under the user attribute characteristic dimension, and the provider federal semi-model outputs a purchasing tendency prediction result of the target long-term risk under the corresponding user attribute characteristic dimension.

In an alternative of this embodiment, combinations with each of the alternatives of one or more of the embodiments described above are possible. Determining a target user for making at least two target long-term risk recommendations may include: determining a plurality of users who purchase a preset short-term risk under a second time window; the first time window is not overlapped with the second time window, and the time point of the first time window is earlier than that of the second time window; and extracting target users needing target long-term risk recommendation from the users purchasing preset short-term risks. For example, 6 short-term risk categories are determined, and an insurance user list of the 6 short-term risk categories is extracted to obtain the target user.

In an alternative of this embodiment, combinations with each of the alternatives of one or more of the embodiments described above are possible. Referring to fig. 3 and 4, determining a target application tendency prediction result of a target user on a target long-term risk by performing purchase tendency prediction on the target long-term risk through an application party federal half model generated by a data application party and a provider federal half model generated by a data provider respectively may include the following steps a 1-A3:

and step A1, determining a first purchasing tendency prediction result of the target user on the target long-term risk based on the application party federal semi-model generated by the data application party.

Step A2, obtaining a second purchasing tendency prediction result of the target user for the target long-term risk from the data provider; the second purchasing propensity prediction result is determined based on a provider federal semi-model generated by the data provider.

And A3, aggregating the first purchasing tendency prediction result and the second purchasing tendency prediction result to obtain a target insurance tendency prediction result of the target user on the target long-term risk.

Referring to fig. 3, determining a first purchasing propensity prediction result of a target user for a target long-term risk based on an application side federal semi-model generated by a data application side may include the following operations: inquiring the purchasing behavior characteristics and insurance product characteristics of the user currently associated with the target user for the insurance product from the local data of the data application party; and inputting the related characteristics of the insurance product purchasing behaviors of the user and the characteristics of the insurance product into an application party federal half model generated by a data application party, and outputting a first purchasing tendency prediction result of the target user on the target long-term risk under the dimension of the model input characteristics.

Referring to fig. 4, the input of the provider federal half model is the user attribute characteristics currently associated with the target user queried in the local data of the data provider, and the output of the provider federal half model is the second purchasing tendency prediction result of the target user on the target long-term risk under the dimension of the user attribute characteristics input by the model. In order to ensure the data security of the second purchasing tendency prediction result, homomorphic encryption processing can be carried out on the second dangerous type insurance prediction result, and the homomorphic encrypted second dangerous type insurance prediction result is further sent to the data application party.

S130, recommending target long-term dangerous seed products to the target users according to the target insurance tendency prediction results of the target users on the target long-term dangerous seeds.

In an alternative of this embodiment, combinations with each of the alternatives of one or more of the embodiments described above are possible. Referring to fig. 6, recommending a target long-term risk product to a target user according to a target application tendency prediction result of the target user for each target long-term risk may include the following steps B1-B2:

b1, according to the insurance application tendency probability of the target user to the target long-term dangerous species included in the target insurance application tendency prediction results, performing descending order arrangement on the target insurance application tendency prediction results; the target application tendency prediction result comprises a target user ID, a target long-term risk species ID and an application tendency probability value.

And step B2, recommending products of long-term dangerous varieties to the target user according to the descending ranking results of the prediction results of the target application tendency.

Referring to fig. 6, a federal model is used for marketing of risk categories of a user group to be predicted, purchasing users of short-term risk categories can be taken as a target user group, purchasing tendencies of the user group to two target long-term risk categories are predicted, for example, a user list of 6 short-term risk categories purchased in a second time window is taken, 103029 users are used in total, purchasing tendencies of all users to the two target long-term risk categories are predicted, purchasing tendencies of the target long-term risk categories are sorted in a descending order, and product recommendation of the long-term risk categories is performed on the target users according to descending order sorting results of prediction results of target insurance application tendencies.

Alternatively, referring to fig. 6, the recommendation of the product of the long-term risk category to the target user according to the descending ranking of the prediction results of the respective target application tendency may include the following operations: selecting a plurality of TOP-n dangerous type product recommendation forms which are ranked in the front and have different numbers of users from the target application tendency prediction results according to the descending ranking results of the target application tendency prediction results; aiming at each TOP-n dangerous type product recommendation form, calculating the target long-term dangerous type application hit rate in the TOP-n dangerous type product recommendation form; and recommending the products of the long-term dangerous seeds to the target user according to the target user proportion in the TOP-n dangerous seed product recommendation list and the target long-term dangerous seed application hit rate in the corresponding TOP-n dangerous seed product recommendation list. For example, take top n (as marketing manifest) and calculate the hit rate of the user's list actually applying the behavior to the two risk categories. The actual insurance application behavior of the user list to the two risk types is shown in the following table:

TABLE 1 user's insurance performance results for two long-term risk categories

Numbering of dangerous species	Name of dangerous species	Number of insuring actions
			BAxx11	xxx New year insurance	32
HDyy11	yyy disease insurance	215

As an alternative, referring to fig. 6, performing long-term risk product recommendation to a target user according to the target user proportion in the TOP-n risk product recommendation form and the target long-term risk exposure hit rate in the corresponding TOP-n risk product recommendation form may include the following operations: determining the insurance tendency probability threshold of the target long-term dangerous seeds according to the target user proportion in the TOP-n dangerous seed product recommendation form and the insurance hit rate of the target long-term dangerous seeds in the corresponding TOP-n dangerous seed product recommendation form; and screening out target insurance tendency prediction results of which the insurance tendency probability is greater than the insurance tendency probability threshold from the target insurance tendency prediction results according to the insurance tendency probability threshold of the target long-term dangerous seeds, and indicating the target user to recommend the products of the long-term dangerous seeds.

By knowing the insurance tendency probability threshold value of the target long-term risk variety, aiming at the long-term risk variety and covering the actual insurance behavior as much as possible with the marketing quantity as less as possible, the target user proportion and the hit rate determination threshold value in the TOP-n risk variety product recommendation form are comprehensively considered according to the following table 2 and the table 3, and a data application party can carry out marketing according to the TOP-n risk variety product recommendation form. For example, marketing HDyy11, with a threshold of 0.37, marketing 0.396 x 103029 users may reach a hit rate of 0.898 actual underwriting.

TABLE 2 marketing plan results for long-term risk categories of HDyy11

Threshold value	0.5	0.49	0.48	0.47	0.46	0.45	0.44	0.43
									Marketing proportion	0.097	0.116	0.128	0.143	0.154	0.168	0.186	0.202
Hit rate	0.186	0.214	0.219	0.256	0.274	0.293	0.298	0.321
									Threshold value	0.42	0.41	0.40	0.39	0.38	0.37	0.36	0.35
Marketing proportion	0.222	0.301	0.317	0.343	0.371	0.396	0.407	0.442
									Hit rate	0.344	0.772	0.814	0.842	0.884	0.898	0.912	0.935
Threshold value	0.34	0.33	0.32	0.31	0.30
									Marketing proportion	0.461	0.494	0.538	0.667	1
Hit rate	0.953	0.958	0.967	0.977	1

TABLE 3 marketing plan results for long-term risk categories of BAxx11

Threshold value	0.5	0.49	0.48	0.47	0.46	0.45	0.44	0.43
									Marketing proportion	0.053	0.055	0.058	0.060	0.065	0.068	0.071	0.076
Hit rate	0.281	0.281	0.281	0.281	0.344	0.344	0.344	0.344
									Threshold value	0.42	0.41	0.40	0.39	0.38	0.37	0.36	0.35
Marketing proportion	0.079	0.084	0.090	0.097	0.103	0.117	0.123	0.130
									Hit rate	0.344	0.344	0.375	0.375	0.438	0.5	0.5	0.5
Threshold value	0.34	0.33	0.32	0.31	0.30
									Marketing proportion	0.144	0.153	0.188	0.312	1
Hit rate	0.5	0.5	0.65	0.969	1

According to the dangerous seed recommendation method provided by the embodiment of the invention, the data application party and the data provider only exchange intermediate results but not original data through a federal learning algorithm, so that the data can be available and invisible to carry out respective model training, and the data combined modeling is realized on the premise of protecting the data privacy; and the data application party and the data provider respectively use the trained half models to predict based on own data, and then collect the respective prediction results to one party to aggregate the prediction results, so that the recommendation of the dangerous seed products is realized.

Fig. 7 is a block diagram of a dangerous seed recommending apparatus provided in an embodiment of the present invention, and the technical solution of this embodiment may be applied to a case of recommending a suitable dangerous seed product to a suitable user. The apparatus can be implemented by software and/or hardware and integrated on any electronic device with network communication function. As shown in fig. 7, the dangerous seed recommending device in the embodiment of the present invention may include the following: a target user determination module 710, a purchasing tendency prediction module 720 and a dangerous goods recommendation module. Wherein:

a target user determination module 710, configured to determine a target user who performs at least two types of target long-term risk recommendation;

the purchasing tendency prediction module 720 is used for respectively predicting purchasing tendency of the target long-term risk species through an application party federal half model generated by a data application party and a provider federal half model generated by a data provider, and determining a target insurance tendency prediction result of the target user on the target long-term risk species;

and the dangerous seed product recommending module 730 is used for recommending the target long-term dangerous seed products to the target user according to the target insurance tendency forecasting result of the target user on each target long-term dangerous seed.

On the basis of the above embodiment, optionally, the application party federal semi-model is obtained by longitudinally federally learning, by a data application party, insurance product purchasing behavior characteristics and insurance product characteristics based on a user under a locally provided first time window under a federal learning framework and user attribute characteristics under the first time window provided by a joint data provider; correspondingly, the provider federal semi-model is obtained by longitudinally and federally learning the insurance product purchasing behavior characteristics and the insurance product characteristics by a user under the first time window provided by the combined data application party on the basis of the user attribute characteristics under the first time window provided by the data provider under the federal learning framework.

On the basis of the above embodiments, optionally, the application side federal half model and the provider side federal half model use a Secureboost federal safety tree-lifting algorithm for model training.

On the basis of the above embodiments, optionally, the application side federal semi-model and the provider side federal semi-model respectively output the prediction result of the purchasing tendency of the target long-term risk under the corresponding model input feature dimension.

On the basis of the foregoing embodiment, optionally, the target user determining module 710 includes:

determining a plurality of users who purchase a preset short-term risk under a second time window; the first time window and the second time window are not overlapped, and the time point of the first time window is earlier than that of the second time window;

and extracting target users needing target long-term risk recommendation from the users purchasing preset short-term risks.

On the basis of the above embodiment, optionally, the purchasing tendency prediction module 720 includes:

determining a first purchasing tendency prediction result of a target user on the target long-term risk based on an application party federal half model generated by a data application party;

acquiring a second purchasing tendency prediction result of the target user on the target long-term risk from the data provider; wherein the second purchasing propensity prediction result is determined based on a provider federal semi-model generated by a data provider;

and aggregating the first purchasing tendency prediction result and the second purchasing tendency prediction result to obtain a target insurance tendency prediction result of the target user on the target long-term dangerous species.

On the basis of the above embodiment, optionally, the second risk insurance prediction result is subjected to homomorphic encryption processing.

On the basis of the foregoing embodiment, optionally, determining a first purchasing tendency prediction result of the target user on the target long-term risk based on an application side federal semi-model generated by the data application side, includes:

inquiring the purchasing behavior characteristics and insurance product characteristics of the user currently associated with the target user for the insurance product from the local data of the data application party;

and inputting the related characteristics of the insurance product purchasing behaviors of the user and the characteristics of the insurance product into an application party federal half model generated by a data application party, and outputting a first purchasing tendency prediction result of the target user on the target long-term risk under the dimension of the model input characteristics.

On the basis of the above embodiment, optionally, the input of the provider federal half model is a user attribute feature currently associated with a target user queried in local data of a data provider, and the output of the provider federal half model is a second purchasing tendency prediction result of the target user on a target long-term risk under the dimension of the user attribute feature input by the model.

On the basis of the foregoing embodiment, optionally, recommending a target long-term risk product to the target user according to a target application tendency prediction result of the target user for each target long-term risk, includes:

according to the insurance tendency probability of the target user to the target long-term dangerous species included in the target insurance tendency prediction results, performing descending arrangement on each target insurance tendency prediction result; the target application tendency prediction result comprises a target user ID, a target long-term dangerous species ID and an application tendency probability value;

and recommending products of long-term dangerous varieties to the target user according to the descending ranking result of the prediction results of the target application tendency.

On the basis of the above embodiment, optionally, the recommending long-term risk products to the target user according to the descending ranking result of the prediction results of the respective target application tendency includes:

selecting a plurality of TOP-n dangerous type product recommendation forms which are ranked in the front and have different numbers of users from the target application tendency prediction results according to the descending ranking results of the target application tendency prediction results;

aiming at each TOP-n dangerous type product recommendation form, calculating the target long-term dangerous type application hit rate in the TOP-n dangerous type product recommendation form;

and recommending the products of the long-term dangerous seeds to the target user according to the target user proportion in the TOP-n dangerous seed product recommendation list and the target long-term dangerous seed application hit rate in the corresponding TOP-n dangerous seed product recommendation list.

On the basis of the above embodiment, optionally, performing long-term risk product recommendation to the target user according to the target user proportion in the TOP-n risk product recommendation form and the target long-term risk application hit rate in the corresponding TOP-n risk product recommendation form, includes:

determining the insurance tendency probability threshold of the target long-term dangerous seeds according to the target user proportion in the TOP-n dangerous seed product recommendation form and the insurance hit rate of the target long-term dangerous seeds in the corresponding TOP-n dangerous seed product recommendation form;

and screening out target insurance tendency prediction results of which the insurance tendency probability is greater than the insurance tendency probability threshold from the target insurance tendency prediction results according to the insurance tendency probability threshold of the target long-term dangerous seeds, and indicating the target user to recommend the products of the long-term dangerous seeds.

The dangerous seed recommending device provided by the embodiment of the invention can execute the dangerous seed recommending method provided by any embodiment of the invention, has corresponding functions and beneficial effects of executing the dangerous seed recommending method, and the detailed process refers to the related operation of the dangerous seed recommending method in the embodiment.

Fig. 8 is a schematic structural diagram of an electronic device provided in an embodiment of the present invention. As shown in fig. 8, the electronic device provided in the embodiment of the present invention includes: one or more processors 810 and storage 820; the processor 810 in the electronic device may be one or more, and fig. 8 illustrates one processor 810 as an example; storage 820 is used to store one or more programs; the one or more programs are executed by the one or more processors 810, such that the one or more processors 810 implement the breeding risk recommendation method of any of the embodiments of the present invention.

The electronic device may further include: an input device 830 and an output device 840.

The processor 810, the storage device 820, the input device 830 and the output device 840 in the electronic apparatus may be connected by a bus or other means, and fig. 8 illustrates an example of connection by a bus.

The storage 820 in the electronic device is used as a computer-readable storage medium for storing one or more programs, which may be software programs, computer-executable programs, and modules, such as program instructions/modules corresponding to the risk recommendation method provided in the embodiments of the present invention. The processor 810 executes various functional applications and data processing of the electronic device by executing software programs, instructions and modules stored in the storage 820, so as to implement the method for recommending dangerous species in the above method embodiments.

The storage device 820 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to use of the electronic device, and the like. Further, storage 820 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid state storage device. In some examples, storage 820 may further include memory located remotely from processor 810, which may be connected to devices over a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The input device 830 may be used to receive input numeric or character information and generate key signal inputs related to user settings and function control of the electronic apparatus. The output device 840 may include a display device such as a display screen.

And, when the one or more programs included in the electronic device are executed by the one or more processors 810, the programs perform the following operations:

Of course, it will be understood by those skilled in the art that when the one or more programs included in the electronic device are executed by the one or more processors 810, the programs may also perform operations related to the risk recommendation method provided in any embodiment of the present invention.

An embodiment of the present invention provides a computer-readable storage medium having stored thereon a computer program, which when executed by a processor, is configured to perform a method for recommending dangerous seeds, the method including:

Optionally, the program, when executed by the processor, may be further configured to perform a method for recommending seeds provided in any embodiment of the present invention.

Computer storage media for embodiments of the invention may employ any combination of one or more computer-readable media. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a Read Only Memory (ROM), an Erasable Programmable Read Only Memory (EPROM), a flash Memory, an optical fiber, a portable CD-ROM, an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. A computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

A computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take a variety of forms, including, but not limited to: an electromagnetic signal, an optical signal, or any suitable combination of the foregoing. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, Radio Frequency (RF), etc., or any suitable combination of the foregoing.

Computer program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).

In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present invention and the technical principles employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A dangerous seed recommendation method, performed by a data application side, the method comprising:

2. The method according to claim 1, wherein the application party federal semi-model is obtained by longitudinally federally learning by a data application party under a federal learning framework based on characteristics of insurance product purchasing behavior and characteristics of insurance products by a user under a first time window provided locally, and user attribute characteristics under the first time window provided by a joint data provider;

correspondingly, the provider federal semi-model is obtained by longitudinally and federally learning the insurance product purchasing behavior characteristics and the insurance product characteristics by a user under the first time window provided by the combined data application party on the basis of the user attribute characteristics under the first time window provided by the data provider under the federal learning framework.

3. The method of claim 2, wherein the application side federated half-model and the provider side federated half-model are model trained using a Secureboost federated safety lifting tree algorithm.

4. The method according to claim 2, wherein the application side federal semi-model and the provider side federal semi-model respectively output the prediction results of the purchasing tendency of the target long-term risk species under the corresponding model input feature dimension.

5. The method of claim 2, wherein determining a target user who makes at least two target long-term risk recommendations comprises:

6. The method according to claim 4, wherein the determining the target application tendency prediction result of the target user on the target long-term risk species by performing purchase tendency prediction on the target long-term risk species through an application party federal semi-model generated by a data application party and a provider federal semi-model generated by a data provider comprises:

7. The method of claim 6, wherein the second risk application prediction result is homomorphic encrypted.

8. The method of claim 6, wherein determining a first buying tendency prediction outcome for the target user for the target long-term risk based on an application side federal semi-model generated by a data application side comprises:

9. The method according to claim 6, wherein the input of the provider federal semi-model is the user attribute characteristics currently associated with the target user queried from the local data of the data provider, and the output of the provider federal semi-model is the second purchasing tendency prediction result of the target user on the target long-term risk under the dimension of the user attribute characteristics input by the model.

10. The method of claim 1, wherein recommending the target long-term risk category product to the target user based on the target application tendency prediction result of the target user for each target long-term risk category comprises:

11. The method of claim 10, wherein recommending long-term risk products to the target user based on the descending ranking of the forecasted target underwriting trends comprises:

12. The method of claim 11, wherein recommending long-term risk products to the target user based on the target user proportion in the TOP-n risk product recommendation form and the target long-term risk application hit rate in the corresponding TOP-n risk product recommendation form comprises:

13. An apparatus for recommending dangerous seeds, configured on a data application side, the apparatus comprising:

14. An electronic device, comprising:

one or more processors;

storage means for storing one or more programs;

when executed by the one or more processors, cause the one or more processors to implement the ethnic risk recommendation method of any of claims 1-12.

15. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the risk recommendation method according to any one of claims 1-12.