Business recommendation method, device, related equipment and storage medium

Abstract

The present invention discloses a service recommendation method, device, server, client device and storage medium. The method includes: obtaining a credit parameter of a first user; the credit parameter represents the credit ability of the first user; using the obtained credit parameter, determining the credit service that the first user can use; using the first data and the second data of the first user, determining the credit service mode preference of the first user; and using the first data and the second data to determine the credit service demand of the first user; the first data represents the credit service historical behavior of the first user; the second data represents the historical behavior of the first user in the communication operator; using the determined credit service that the first user can use, the credit service mode preference, and the credit service demand, recommending credit services to the first user.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a service recommendation method, device, related equipment and storage medium. Background Art

[0002] With the development of Internet technology and communication technology and the construction of the social credit system, various service scenarios based on credit scores have gradually developed to provide users with various services. Communication operators have huge data assets and can better evaluate the credit ability of users. Based on the credit scores evaluated by communication operators, various credit application services can be developed, such as credit purchase of mobile phones, credit hotel stays, credit car rentals, credit communications (no downtime), credit book borrowing, credit number loans, etc. In recent years, with the gradual maturity of credit assessment technology and credit application systems, the credit service field and credit business scenarios have become more and more abundant.

[0003] In the related art, the recommendation accuracy of the credit service recommendation technology is poor, which leads to a poor user experience. Summary of the invention

[0004] To solve the related technical problems, the embodiments of the present invention provide a service recommendation method, apparatus, related equipment and storage medium.

[0005] The technical solution of the embodiment of the present invention is achieved as follows:

[0006] The embodiment of the present invention provides a service recommendation method, which is applied to a server and includes:

[0007] Acquiring a credit parameter of the first user; the credit parameter represents the credit ability of the first user;

[0008] Determining a credit service that the first user can use by using the acquired credit parameter and the identity information of the first user;

[0009] Determine the credit service mode preference of the first user by using the first data and the second data of the first user; and determine the credit service demand of the first user by using the first data and the second data; the first data represents the credit service historical behavior of the first user; the second data represents the historical behavior of the first user in the communication operator;

[0010] The determined credit services available to the first user, the credit service model preference, and the credit service needs are used to recommend credit services to the first user.

[0011] In the above solution, the step of using the acquired credit parameter and the identity information of the first user to determine the credit service that the first user can use includes:

[0012] For each of at least two types of credit services in the database, determine from the corresponding type of credit services a service that matches the acquired credit parameters and a corresponding service level, or determine from the corresponding type of credit services a service that matches the identity information of the first user and a corresponding service level.

[0013] In the above solution, the using the first data and the second data of the first user to determine the credit service mode preference of the first user includes:

[0014] Performing a first feature extraction on the first data and the second data to obtain a first extraction result, and performing a second feature extraction on the first data and the second data to obtain a second extraction result; the first extraction result represents the behavior of the first user; and the second extraction result represents the time series of the first user;

[0015] Using the first extraction result in combination with the first model, evaluating the credit service mode preference of the first user to obtain a first credit service mode preference result;

[0016] Using the second extraction result and a second model, evaluating the credit service mode preference of the first user to obtain a second credit service mode preference result;

[0017] The credit service mode preference of the first user is determined by using the first credit service mode preference result and the second credit service mode preference result.

[0018] In the above solution, the step of determining the credit service mode preference of the first user by using the first credit service mode preference result and the second credit service mode preference result includes:

[0019] Normalizing the first credit service model preference result to obtain a first processing result; and normalizing the second credit service model preference result to obtain a second processing result;

[0020] The credit service mode preference of the first user is determined by using the first processing result and the second processing result.

[0021] In the above solution, the using the first data and the second data to determine the credit service demand of the first user includes:

[0022] Extracting features from the first data and the second data to obtain user demand features;

[0023] When the obtained user demand characteristics meet the first condition, the obtained user demand characteristics are input into a deep learning model to obtain the credit business demand of the first user; the deep learning model is established based on a latent Dirichlet allocation-long short-term memory network (LDA-LSTM) algorithm using training samples;

[0024] In the case where the obtained user demand feature meets the second condition, the obtained user demand feature is input into the expert rule recommendation model to obtain the credit business demand of the first user; the expert rule recommendation model is constructed by using training samples to construct demand labels and is established based on the constructed demand labels;

[0025] The first condition is different from the second condition.

[0026] In the above scheme, when the credit business demand of the first user obtained by using the deep learning model does not match the recommended item filtering rules corresponding to the deep learning model, the obtained user demand characteristics are input into the expert rule recommendation model to obtain the credit business demand of the first user.

[0027] In the above scheme, when the credit business demand of the first user obtained by using the expert rule recommendation model does not match the recommendation item filtering rule corresponding to the expert rule recommendation model, the obtained user demand characteristics are input into the random strategy recommendation model to obtain the credit business demand of the first user; the random strategy recommendation model is obtained based on the sorting result of the recommendation probability.

[0028] In the above solution, using the determined credit services available to the first user, the credit service model preference, and the credit service needs, recommending credit services to the first user includes:

[0029] Merging the scores corresponding to each credit service among the credit services available to the first user, the scores corresponding to each credit service among the credit services corresponding to the credit service mode preference, and the scores corresponding to each credit service in the credit service demand to obtain a fusion result;

[0030] A credit service recommendation is made to the first user based on the fusion result.

[0031] In the above scheme, the credit service model preference is matched with the service model of the credit business in the database for similarity, and based on the similarity matching result, the score corresponding to each credit business in the credit business corresponding to the credit service model preference is determined.

[0032] In the above solution, when recommending a credit service to the first user, the method includes:

[0033] Sending credit service recommendation information to the client device corresponding to the first user; the service recommendation information includes credit service and corresponding grade information.

[0034] In the above scheme, the method further comprises:

[0035] Acquiring user data generated by the first user for a first credit service; the first credit service is a credit service selected by the first user after a credit service recommendation is made to the first user;

[0036] Use the generated user data to update the corresponding model.

[0037] The embodiment of the present invention further provides a service recommendation method, which is applied to a client device, comprising:

[0038] Receiving credit service recommendation information sent by a server; the service recommendation information includes credit service and corresponding grade information;

[0039] Based on the received credit service recommendation information, a credit service recommendation is presented; the presented content includes the credit service and the corresponding grade information.

[0040] In the above scheme, the method further comprises:

[0041] Send the identity information of the first user to the server.

[0042] The embodiment of the present invention further provides a service recommendation device, including:

[0043] An acquisition unit, configured to acquire a credit parameter of the first user; the credit parameter represents the credit ability of the first user;

[0044] A first determining unit, configured to determine a credit service that the first user can use by using the acquired credit parameter and the identity information of the first user;

[0045] A second determination unit is used to determine the credit service mode preference of the first user by using the first data and the second data of the first user; and to determine the credit service demand of the first user by using the first data and the second data; the first data represents the credit service historical behavior of the first user; and the second data represents the historical behavior of the first user in the communication operator;

[0046] The third determining unit is used to recommend a credit service to the first user by using the determined credit services available to the first user, the credit service model preference, and the credit service demand.

[0047] The embodiment of the present invention further provides a service recommendation device, including:

[0048] A transceiver unit, configured to receive credit service recommendation information sent by a server; the service recommendation information includes credit services and corresponding grade information;

[0049] The processing unit is used to present a credit service recommendation based on the received credit service recommendation information; the presented content includes the credit service and the corresponding grade information.

[0050] An embodiment of the present invention further provides a server, comprising: a first processor and a first memory for storing a computer program that can be run on the processor, wherein:

[0051] The first processor is used to execute the steps of any of the above-mentioned server-side methods when running the computer program.

[0052] The embodiment of the present invention further provides a client device, comprising: a second processor and a second memory for storing a computer program that can be run on the processor, wherein:

[0053] The second processor is used to execute the steps of any of the above-mentioned methods on the client device side when running the computer program.

[0054] An embodiment of the present invention further provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods on the server side are implemented, or the steps of any of the above-mentioned methods on the client device side are implemented.

[0055] The service recommendation method, apparatus, related equipment and storage medium provided in the embodiments of the present invention obtain the credit parameters of the first user; the credit parameters represent the credit ability of the first user; the credit services available to the first user are determined by using the obtained credit parameters and the identity information of the first user; the credit service mode preference of the first user is determined by using the first data and the second data of the first user; and the credit service demand of the first user is determined by using the first data and the second data; the first data represents the historical credit service behavior of the first user; the second data represents the historical behavior of the first user at the communication operator; the credit service, credit service mode preference and credit service demand that the first user can use are used to recommend the credit service to the first user, and the credit service recommendation is made to the user through a multi-level evaluation model of credit ability, credit service mode preference and credit service demand, which can effectively improve the accuracy of credit service recommendation and improve user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A schematic diagram of a method flow chart for service recommendation in an embodiment of the present invention;

[0057] Figure 2 A schematic diagram of an evaluation mechanism of a credit business evaluation and recommendation technical solution according to an application embodiment of the present invention;

[0058] Figure 3 This is a schematic diagram of the structure of a credit business preference, recommendation and display system according to an application embodiment of the present invention;

[0059] Figure 4 A schematic diagram of a process for recommending credit services in an application embodiment of the present invention;

[0060] Figure 5 A schematic diagram of a credit service process that can be used by a user in evaluating an application embodiment of the present invention;

[0061] Figure 6 This is a schematic diagram of the mapping relationship between credit service level and score in an application embodiment of the present invention;

[0062] Figure 7 A schematic diagram of a model flow for evaluating credit service model preferences is constructed for an application embodiment of the present invention;

[0063] Figure 8 A schematic diagram of a process for evaluating a credit business willingness evaluation score according to an application embodiment of the present invention;

[0064] Fig. 9 A schematic diagram of a classifier processing flow for a service mode according to an application embodiment of the present invention;

[0065] Fig.10 This is a schematic diagram of the prediction result of the credit installment amount according to an application example of the present invention;

[0066] Fig.11 A schematic diagram of a process for constructing an ARMA-based model for an application embodiment of the present invention;

[0067] Fig.12 A schematic diagram of a mapping process of credit service model preference evaluation results to credit business preferences in an application embodiment of the present invention;

[0068] Fig.13 A schematic diagram of the collaborative linkage recommendation process of the application embodiment of the present invention "expert knowledge rules + deep learning";

[0069] Fig.14 This is a schematic diagram of an LDA-LSTM classifier according to an application example of the present invention;

[0070] Fig.15 A schematic diagram of a feedback mechanism for an application example of the present invention;

[0071] Fig.16 A schematic diagram of a method flow chart of post-evaluation processing in an application example of the present invention;

[0072] Fig.17 A schematic diagram of credit business recommendation results for user A and user B when the relevant technology is used;

[0073] Fig.18 This is a schematic diagram of credit service recommendation results for user A and user B when the solution of an embodiment of the present invention is adopted;

[0074] Fig.19 This is a schematic diagram of the structure of a service recommendation device according to an embodiment of the present invention;

[0075] Fig. 20 This is a schematic diagram of the structure of another service recommendation device according to an embodiment of the present invention;

[0076] Fig.21 This is a schematic diagram of the server structure of an embodiment of the present invention;

[0077] Fig. 22 Schematic diagram of the structure of a client device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0078] The present invention will be described in further detail below in conjunction with the accompanying drawings and embodiments.

[0079] At present, common business recommendation solutions mainly recommend products or services based on user preferences, including content-based recommendation, collaborative filtering-based recommendation, association rule-based recommendation, knowledge-based recommendation, context-based recommendation, etc. These solutions are traditional fixed and indifferent business recommendation solutions.

[0080] At present, there is no personalized recommendation technology solution for credit business recommendation. In the related technology, a recommendation solution for credit business is composed of a client, a credit score query server, and a business party server, and adopts a traditional fixed and indifferent business recommendation solution. In other words, the business recommendation solution is only related to the importance of the business, and has nothing to do with the credit assessment results and demand preferences of individual users. This technical solution is specifically manifested in that after the user performs a credit score query and display on the client, the credit business list obtained by different users is unified and indistinguishable, without matching demand intentions and ability assessment results. In other words, after entering the credit service module, all users obtain the same credit business recommendation, and it is impossible to accurately assess the user's needs and preferences for credit services.

[0081] Therefore, it is necessary to build a personalized recommendation technology solution for credit business to improve user experience.

[0082] On the other hand, the difference between credit business recommendation and common business (commodity, content or service, etc.) recommendation is that: for common business (commodity, content or service, etc.), there is usually no purchase threshold or usage threshold; while for credit business, there are two characteristics: business usage threshold and level differences and credit model differences due to differences in credit ability. Therefore, in addition to evaluating the user's demand for goods, content or services, credit business recommendation is more importantly to accurately and quickly evaluate the user's scope of application of credit business based on credit ability, and accurately evaluate the user's preference for credit service model based on the user's credit behavior.

[0083] Based on this, in various embodiments of the present invention, credit service recommendations are made to users through a multi-level evaluation model of credit ability, credit service model preference, and credit service demand, which can effectively improve the accuracy of credit service recommendations.

[0084] The embodiment of the present invention provides a service recommendation method, such as Figure 1 As shown, the method includes:

[0085] Step 101: Obtain credit parameters of a first user;

[0086] Here, the credit parameter represents the credit ability of the first user.

[0087] Step 102: using the obtained credit parameter and the identity information of the first user, determining the credit service that the first user can use;

[0088] Step 103: using the first data and the second data of the first user, determining the credit service mode preference of the first user; and using the first data and the second data, determining the credit service demand of the first user;

[0089] Here, the first data represents the credit business history behavior of the first user; the second data represents the history behavior of the first user in the communication operator;

[0090] Step 104: Recommend credit services to the first user by using the determined credit services available to the first user, credit service model preferences, and credit service needs.

[0091] It should be noted that, in actual application, the execution order of step 103 and steps 101 and 102 can be partially sequential, that is, the execution order of steps 101 to 103 can be: steps 101, 102, 103, or steps 101, 103, 102, or 103, 101, 102.

[0092] The solution provided in the embodiment of the present invention can be applied to a server; accordingly, the server can send the credit service to the client used by the first user to recommend the credit service to the first user.

[0093] In step 101, in actual application, the credit parameter may be expressed as a credit score. Specifically, the server may query the credit score according to the user identity information (such as user ID) sent by the first user when logging in through the client device, thereby obtaining the credit score of the first user.

[0094] In step 102, in actual application, the identity information of the first user may be the ID of the first user, etc., as long as the identity of the first user can be identified, and this is not limited in the embodiment of the present invention.

[0095] According to the service rules, the credit services may be classified so that the credit services that the first user can use may be determined according to different service evaluation rules.

[0096] Based on this, in one embodiment, using the acquired credit parameter to determine the credit service that the first user can use includes:

[0097] For each of at least two types of credit services in the database, determine from the corresponding type of credit services a service that matches the acquired credit parameters and a corresponding service level, or determine from the corresponding type of credit services a service that matches the identity information of the first user and a corresponding service level.

[0098] That is to say, when determining the credit services that the first user can use from the corresponding types of credit services, for some corresponding types of credit services, the acquired credit parameters are used to match the credit services that the first user can use and the corresponding service levels; for some corresponding types of credit services, the identity information of the first user is used to match the credit services that the first user can use and the corresponding service levels.

[0099] For example, according to business rules, credit services can be divided into first-type credit services and second-type credit services. Among them, the first-type credit services can be the credit services owned by telecom operators, such as credit without service shutdown, credit book borrowing (such as China Mobile Migu Bookstore book borrowing service), credit opening of international roaming, etc. The second-type credit services can be cooperative businesses between telecom operators and other industries, such as credit purchase of mobile phones, credit hotel stays, credit hospitalization, credit cycling, etc.

[0100] For the first type of business, since the credit score is evaluated based on communication big data, this type of business can directly use the credit score to formulate business rules and conditions. Credit business level matching can be performed through credit score threshold matching, that is, in this way, the processing speed of business matching can be effectively improved.

[0101] As for the second type of services, it is necessary to conduct demand and risk assessment on individual users, so this type of services can match the credit services and corresponding service levels available to the first user through the user's identity information.

[0102] In actual application, the first user may be able to use multiple credit services. In order to make subsequent recommendations, multiple credit services need to be scored. The grade corresponding to each credit service can be used to score the credit service, that is, the grade corresponding to each service is mapped to a score. The mapping method can adopt linear mapping, nonlinear mapping, expert scoring method, etc., which is not limited to the embodiment of the present invention.

[0103] The mapping method follows the following principles:

[0104] The higher the level, the higher the score;

[0105] The higher the business importance, the higher the score.

[0106] In step 103, in actual application, the user's credit service mode preference can be determined based on the data generated by the user at the communication operator and the credit business behavior data generated by the credit business. Using multiple data to determine the user's credit service mode preference can improve the accuracy of determining the credit service mode preference.

[0107] Based on this, in one embodiment, using the first data and the second data of the first user to determine the credit service mode preference of the first user includes:

[0108] Performing a first feature extraction on the first data and the second data to obtain a first extraction result, and performing a second feature extraction on the first data and the second data to obtain a second extraction result; the first extraction result represents the behavior of the first user; and the second extraction result represents the time series of the first user;

[0109] Using the first extraction result in combination with the first model, evaluating the credit service mode preference of the first user to obtain a first credit service mode preference result;

[0110] Using the second extraction result and a second model, evaluating the credit service mode preference of the first user to obtain a second credit service mode preference result;

[0111] The credit service mode preference of the first user is determined by using the first credit service mode preference result and the second credit service mode preference result.

[0112] In actual application, the first data and the second data may be preprocessed first, and the first feature extraction and the second feature extraction may be performed using the preprocessed data. The preprocessing may include data cleaning, data integration, data transformation, data reduction, etc.

[0113] The first model may be called a credit service model classifier, that is, a classification model, which is used to predict the probability of a user using a certain credit service model and the specific subcategories under the credit service model.

[0114] The second model can be called a time series model, which is used to predict specific usage indicators of users under a certain credit service model, such as credit usage limit, usage cycle, usage interest, etc.

[0115] The first credit service model preference result and the second credit service model preference result are used to determine the credit service model preference of the first user, and the evaluation results of the two models are integrated, so that a more accurate credit service model preference result can be obtained.

[0116] In actual application, since the first credit service model preference result and the second credit service model preference result are expressed in different forms, in order to make the corresponding indicators of the two comparable, the two can be standardized, or can be understood as normalized.

[0117] Based on this, in one embodiment, determining the credit service mode preference of the first user by using the first credit service mode preference result and the second credit service mode preference result includes:

[0118] Normalizing the first credit service model preference result to obtain a first processing result; and normalizing the second credit service model preference result to obtain a second processing result;

[0119] The credit service mode preference of the first user is determined by using the first processing result and the second processing result.

[0120] In actual application, the credit services corresponding to the credit service model preference of the first user may include multiple ones. For subsequent recommendation, it is necessary to score the credit services based on the credit service model preference, that is, map the credit service model preference to the credit service preference.

[0121] Based on this, in one embodiment, the method may further include:

[0122] The credit service model preference is matched with the service model of the credit business in the database for similarity, and based on the similarity matching result, the score corresponding to each credit business in the credit business corresponding to the credit service model preference is determined.

[0123] In one embodiment, the using the first data and the second data to determine the credit service demand of the first user includes:

[0124] Extracting features from the first data and the second data to obtain user demand features;

[0125] When the obtained user demand characteristics meet the first condition, the obtained user demand characteristics are input into a deep learning model to obtain the credit business demand of the first user; the deep learning model is established based on the LDA-LSTM algorithm using training samples;

[0126] In the case where the obtained user demand feature meets the second condition, the obtained user demand feature is input into the expert rule recommendation model to obtain the credit business demand of the first user; the expert rule recommendation model is constructed by using training samples to construct demand labels and is established based on the constructed demand labels;

[0127] The first condition is different from the second condition.

[0128] In actual application, the user demand refers to the user's demand for specific credit business layers such as purchasing terminals (such as mobile phones, etc.), renting cars, staying in hotels, reading, etc.

[0129] It can be seen from the above description that in the embodiment of the present invention, a hybrid feature mining method of an expert rule recommendation model (which can be understood as expert knowledge rules) combined with a deep learning model is used to determine the user's credit business needs, which can ensure that users with different data conditions and evaluation conditions can obtain the best recommendation results within their own evaluable range.

[0130] When it is determined that the number of the obtained user demand features meets the third condition, it is determined that the obtained user demand features meet the first condition.

[0131] Among them, the third condition characterizes that the number of obtained user demand features is sufficient. In actual application, corresponding feature quantity thresholds can be set for different types of features. When the types of features exceeding the threshold meet certain conditions, it means that the number of user demand features is sufficient (for example, there are five types, and the types of features exceeding the threshold are greater than or equal to three, it is considered that the number of user demand features is sufficient. Of course, there may be other ways, which are not limited in the embodiments of the present invention). The embodiments of the present invention do not limit how to determine whether the number of user demand features is sufficient.

[0132] When it is determined that the number of the obtained user demand features does not satisfy the third condition, it is determined that the obtained user demand features satisfy the second condition.

[0133] In actual application, in order to ensure more accurate recommendation results, when the confidence level of the credit business demand of the first user obtained by one model is not high, the credit business demand of the first user obtained by another model can be used as the determined credit business demand of the first user.

[0134] Based on this, in one embodiment, when the credit business demand of the first user obtained by using the deep learning model does not match the recommendation item filtering rule corresponding to the deep learning model, the obtained user demand characteristics are input into the expert rule recommendation model to obtain the credit business demand of the first user.

[0135] When the credit business demand of the first user is obtained by using the expert rule recommendation model, if the credit business demand of the first user obtained by using the expert rule recommendation model does not match the recommendation item filtering rule corresponding to the expert rule recommendation model, the obtained user demand characteristics are input into the random strategy recommendation model to obtain the credit business demand of the first user; the random strategy recommendation model is obtained based on the sorting result of the recommendation probability.

[0136] In actual application, the credit service demand corresponding to the first user may include multiple services. For subsequent recommendation, each credit service in the determined credit service demand has a corresponding score.

[0137] After the above processing, the specific implementation of step 104 may include:

[0138] Merging the scores corresponding to each credit service among the credit services available to the first user, the scores corresponding to each credit service among the credit services corresponding to the credit service mode preference, and the scores corresponding to each credit service in the credit service demand to obtain a fusion result;

[0139] A credit service recommendation is made to the first user based on the fusion result.

[0140] The fusion processing may be: for a credit service, weighting the score corresponding to the available credit services, the score corresponding to the credit services corresponding to the credit service model preference, and the score corresponding to the credit service demand to obtain the score of the credit service, and so on to obtain the scores of each credit service, and sort them according to the scores (from high to low). When recommending, select the first N credit services for recommendation, and the value of N is determined according to needs.

[0141] In actual application, in order to achieve differentiated credit service recommendations for different users, the level corresponding to the user can be further recommended when making credit service recommendations.

[0142] Based on this, in one embodiment, when recommending a credit service to the first user, the method includes:

[0143] Sending credit service recommendation information to the client device corresponding to the first user; the service recommendation information includes credit service and corresponding grade information.

[0144] In practical applications, in order to make the recommendation results more and more accurate, a feedback loop and model iteration method can be used for optimization. That is, the feedback results of the recommendation results are used to optimize (update) the models used in the above steps.

[0145] Based on this, in one embodiment, the method may further include:

[0146] Acquiring user data generated by the first user for a first credit service; the first credit service is a credit service selected by the first user after a credit service recommendation is made to the first user;

[0147] Use the generated user data to update the corresponding model.

[0148] Correspondingly, an embodiment of the present invention provides a service recommendation method, which is applied to a client device, including:

[0149] Receiving credit service recommendation information sent by a server; the service recommendation information includes credit service and corresponding grade information;

[0150] Based on the received credit service recommendation information, a credit service recommendation is presented; the presented content includes the credit service and the corresponding grade information.

[0151] In one embodiment, the method may further include:

[0152] Send the identity information of the first user to the server.

[0153] Here, in actual application, after the first user logs in to the client device, the client device may send the identity information of the first user to the server.

[0154] The client device may be a mobile phone, a tablet computer, etc.

[0155] The service recommendation method provided in the embodiment of the present invention obtains the credit parameters of the first user; the credit parameters represent the credit ability of the first user; the credit services available to the first user are determined by using the obtained credit parameters and the identity information of the first user; the credit service mode preference of the first user is determined by using the first data and the second data of the first user; and the credit service demand of the first user is determined by using the first data and the second data; the first data represents the credit service historical behavior of the first user; the second data represents the historical behavior of the first user in the communication operator; the credit service, credit service mode preference and credit service demand that the first user can use are used to recommend the first user a credit service, and the credit service recommendation is made to the user through a multi-level evaluation model of credit ability, credit service mode preference and credit service demand, which can effectively improve the accuracy of credit service recommendation and improve user experience.

[0156] The present invention is further described in detail below in conjunction with application examples.

[0157] In the application embodiment of the present invention, Figure 2 As shown in the figure, the technical solution for credit business evaluation and recommendation should be generally divided into three-tier evaluation mechanisms: the first is the user's credit ability assessment, which evaluates the scope and level of credit business that the user can enjoy based on the user's credit score information; the second is the user's credit service model preference assessment, which evaluates the user's preference for the credit service model, willingness to use the credit limit, etc.; the third is the user's business (goods / services) demand assessment, which evaluates the user's demand for the goods / services involved in the credit business.

[0158] In the application embodiment of the present invention, Figure 3 As shown, the credit service preference, recommendation and display system (also referred to as the credit service personalized recommendation and display system) includes: a credit application client, a credit evaluation server, M business party servers, and a demand evaluation server.

[0159] The credit application client provides users with functions such as checking credit scores, browsing and applying for credit services, and using credit services. The client mainly includes: user login module, score display module, and service display module. When users browse and apply for credit services through the client, they need to use the functions of the above three modules.

[0160] The credit assessment server provides functions such as credit score query, credit business capability assessment and matching, credit service model preference assessment, credit business demand assessment, and comprehensive recommendation. The credit assessment server mainly includes a credit score query module, a credit business capability assessment and matching module, a credit service model preference assessment module, a business demand assessment module, and a credit business recommendation module.

[0161] The business server is the server where all kinds of credit services are located, such as the services of provincial companies of telecommunications operators, the services of various professional companies, and the services of external partners such as financial institutions. The business server mainly includes the credit business evaluation module and the credit business behavior feedback module.

[0162] The business demand assessment server provides the function of evaluating users' business demand labels for goods / services based on user behavior / terminal / communication and other dimensional data. The business demand assessment server mainly includes the business demand assessment module.

[0163] based on Figure 3 The system shown in FIG. 1 shows a process for recommending credit services in this application embodiment. Figure 4 As shown, including:

[0164] Step 1: After the user logs in to the client, the client sends the user identity information to the credit assessment server;

[0165] Here, the user logs in through the user login module of the client and sends the user identity information to the credit evaluation server.

[0166] The user identity information is the user ID.

[0167] Step 2: After receiving the user ID, the credit assessment server performs a credit score query based on the user ID to obtain the user's credit score;

[0168] Here, the credit score query module of the credit evaluation server performs a credit score query based on the user ID to obtain the user's credit score.

[0169] Step 3: The credit assessment server assesses the service matching scope and level, that is, assesses the credit services that the user can use and the corresponding level;

[0170] Specifically, the credit assessment server performs condition matching on the first category of business according to the credit score query result to obtain a first category of business matching result;

[0171] The credit assessment server performs condition matching on the second type of business according to the user ID, and obtains a matching result for the second type of business;

[0172] The credit evaluation server performs credit business matching management, that is, combines the matching results of the first type of business and the second type of business to obtain all business matching results of the user and the corresponding evaluation score (which can be called the credit business capability evaluation score) Scrd_ab.

[0173] This step is performed by the credit business capability assessment and matching module of the credit assessment server.

[0174] Step 4: The credit evaluation server evaluates the user's credit service model preference and obtains the credit service model preference evaluation result (i.e., the credit service willingness evaluation score) Scrd_ds;

[0175] Here, this step is performed by the credit service mode preference evaluation module of the credit evaluation server.

[0176] Step 5: The credit evaluation server evaluates the user's credit service needs; queries the user's demand tag evaluation results based on the credit service matching results to obtain the user's demand evaluation results for specific services (which can be called credit service demand evaluation scores) Sbsn_ds;

[0177] Here, this step is performed by the business demand assessment module of the credit assessment server.

[0178] Step 6: The credit assessment server establishes a multi-layer fusion recommendation mechanism based on the credit business matching results, the credit service model preference assessment results, and the user demand assessment results, conducts a comprehensive personalized recommendation assessment, and feeds back the assessment results to the client;

[0179] In other words, make comprehensive recommendations for credit business.

[0180] Here, this step is performed by the credit service recommendation module of the credit evaluation server.

[0181] Step 7: The client makes corresponding personalized recommendations and displays based on the evaluation results of the credit evaluation server.

[0182] Here, this step is performed by the score display module and the business display module of the client.

[0183] Some of the steps are described in detail below.

[0184] In step 3, a credit capability assessment is performed, that is, an assessment of the credit services that the user can use. Figure 5 As described above, the user ID and credit score are matched with the credit service capability matching library to obtain the credit services and corresponding levels that the user can use; then, the obtained credit services and corresponding levels are used to evaluate the credit service capability to obtain the user's credit service capability score, that is, the credit evaluation score.

[0185] The process of matching in the credit business capability matching library includes: matching in both the first category of business and the second category of business.

[0186] Credit business is divided into two types of business according to the business rules: first-class business and second-class business.

[0187] The first type of business is usually the telecom operator's own business, such as credit without shutdown, credit borrowing (such as China Mobile Migu Bookstore borrowing business), credit activation of international roaming, etc. Since the credit score is based on the evaluation of telecom big data, the first type of business can directly use the credit score to formulate business rules and conditions. The evaluation rule corresponding to the first type of business is to judge the business level through the credit score condition. For example, if the credit score is greater than a certain value, you can enjoy a deposit-free service. For the first type of business, the credit evaluation server matches the credit business level through the credit score threshold, which can effectively improve the processing speed of business matching.

[0188] Exemplarily, the evaluation rule database D1 for the first type of business (i.e., the credit business capability matching database) is shown in Table 1:

[0189] Business ID Business Name Judgment conditions (credit score) Service Level Level Description A1 No credit interruption [350,550) 0 No credit, no downtime qualification A1 No credit interruption [550,650) 1 100 yuan overdraft limit A1 No credit interruption [650,750) 2 300 yuan overdraft limit A2 Credit loan [350,450) 0 No deposit exemption qualification A2 Credit loan [450,750) 1 Borrow books without deposit

[0190] Table 1

[0191] When matching, the first type of business servers including M1 business servers evaluate the rules of their respective corresponding businesses and feed back the evaluation rules to the credit evaluation server, thereby supporting the credit business matching process in the credit evaluation server.

[0192] When the credit evaluation server interacts with the first-category business server, an offline processing method can be used. Specifically, an offline processing solution is adopted, and a first-category business evaluation rule database D1 is set in the credit evaluation server to store the corresponding evaluation rules of the credit score and various credit businesses. The N1 business party servers corresponding to the first-category business periodically synchronize the evaluation rules to the credit evaluation server. In this way, the credit evaluation server can query the user's credit score in D1 to obtain the credit business matching result of the user.

[0193] Then, for a user, after the above processing, the following processing results are obtained:

[0194] R1=[(A1, ra1), (A2, ra2,...].

[0195] The second type of business is usually a cooperative business between telecommunications operators and other industries, such as credit purchase of mobile phones, credit hotel stays, credit hospitalization, credit cycling, etc. In addition to judging by credit score conditions, the second type of business also needs to conduct a joint evaluation of users through personal data and partner data to make a judgment on the service level. This type of business will further evaluate users who meet the credit score conditions: based on the personal data of the communication operator and the data of the partner (such as financial institutions, hotels and other cooperative institutions), the needs and risks of individual users are evaluated for this business, and each user is evaluated through each business evaluation model to obtain their service level and credit limit. Among them, the specific implementation of the business evaluation is not limited in the embodiments of the present invention. Therefore, the second type of business matches the credit business level through the user ID.

[0196] Exemplarily, the evaluation rule database D2 for the second type of business (i.e., the credit business capability matching database) is shown in Table 2:

[0197]

[0198] Table 2

[0199] In Table 2, the credit purchase service (B1) is taken as an example. First, users with a credit score of 450 or above can enjoy the credit purchase service:

[0200] The credit score of user with ID U1 (400 points) does not meet the threshold, so he cannot enjoy the credit purchase service and his service level is 0; the credit scores of users with ID U2 and U3 (600 points and 700 points respectively) meet the threshold, and their credit limits are evaluated through the credit purchase evaluation model. They are given credit limits of 1,000 yuan and 2,000 yuan respectively, and their levels are 1 and 2 respectively.

[0201] When matching, the second type of business server includes M2 business servers, which perform rule evaluation and user matching for their respective businesses, and feed back the user matching results to the credit evaluation server, thereby supporting the credit business matching process in the credit evaluation server.

[0202] When the credit assessment server interacts with the second type of business server, it can choose an online processing solution or an offline processing method based on system performance.

[0203] If online processing is adopted, the credit assessment server sends the identity ID and credit score query results to the M2 business party servers corresponding to the second category of business respectively; each of the M2 credit business servers performs condition matching for the second category of business, that is, according to the sent identity ID and credit score query results, it matches the conditions for its own business and feeds back the matching results to the credit assessment server.

[0204] If an offline processing solution is adopted, a second-category business evaluation rule database D2 is set in the credit evaluation server to store the matching results of users and various credit services. The credit evaluation server regularly sends the user ID and credit score to the M2 business party servers corresponding to the second-category business. The business party server performs business evaluation based on the user ID and credit score, and regularly synchronizes the business evaluation results to the credit evaluation server. Therefore, by querying in the database D2 according to the user ID, the credit business matching results of the user can be obtained. This solution can effectively improve the recommendation processing speed.

[0205] Then, for a user, after the above processing, the following processing results are obtained:

[0206] R2=[(B1, rb1), (B2, rb2,...].

[0207] After obtaining the matching results, it is necessary to conduct a comprehensive evaluation of the credit business based on the obtained matching results R1 and R2, and map the service level to the credit evaluation score indicator for the credit business, that is, to obtain the user's credit business capability score, which specifically includes:

[0208] First, we obtain the credit business matching results of the customer from the first and second types of business. The comprehensive matching results for a certain user are:

[0209] R=[R1, R2]=[(A1, ra1), (A2, rb2),..., (B1, rb1), (B2, rb2),...]

[0210] Secondly, credit business grade score mapping is performed, that is, the evaluation grade results of each credit business are mapped to scores for subsequent personalized recommendation processing.

[0211] The mapping method is set in combination with the distribution of each credit business level. The mapping method can adopt linear mapping, nonlinear mapping, expert scoring method, etc., which is not limited in the embodiment of the present invention.

[0212] The mapping method follows the following principles:

[0213] The higher the level, the higher the score;

[0214] The higher the business importance, the higher the score.

[0215] The mapping functions of different credit businesses are as follows:

[0216] Scrd_ab A1 =f A1 (r)

[0217] Scrd_ab A2 =f A2 (r)

[0218] Scrd_ab B1 =f B1 (r)

[0219] Scrd_ab B2 =f B2 (r)

[0220] For example, a mapping method is implemented as shown in Table 3 and Figure 6 shown.

[0221] Business level Level 0 Level 1 Level 2 Level 3 Level 4 Business A1 0 20 40 — — Business A2 0 30 — — — Business B1 0 15 25 40 — Business B2 0 10 20 30 40

[0222] Table 3

[0223] Among them, Figure 6 In the example, -- indicates business A1, - indicates business A2, Indicates business B1, Indicates business B2.

[0224] For a certain user, after the above processing, the processing results are as follows:

[0225] Scrd_ab=[(A1,Scrd_ab A1 ), (A2, Scrd_ab A2 ), (B1, Scrd_ab B1 ), (B2, Scrd_ab B2 )]

[0226] For user U1: Scrd_ab = [(A1, 0), (A2, 0), (B1, 0), (B2, 10)]

[0227] For user U2: Scrd_ab = [(A1, 20), (A2, 30), (B1, 15), (B2, 20)]

[0228] For user U3: Scrd_ab = [(A1, 40), (A2, 30), (B1, 25), (B2, 30)]

[0229] From the above description, it can be seen that the credit capacity assessment method provided in the embodiment of the present invention can output the assessment results to the comprehensive recommendation model in a timely and rapid manner by optimizing the credit business matching processing steps, that is, it can effectively improve the accuracy of credit business recommendations; by optimizing the system linkage processing steps (business classification and matching database optimization, and optimization of the interaction mechanism between the credit assessment server and the business party server), the credit business recommendation speed is effectively improved.

[0230] Based on credit, we can provide users with a variety of credit service models in traditional business models, such as no deposit model, no deposit model, enjoy first and pay later model, consumption installment model, fast review-free model, credit reservation model, etc. Based on the user's usage behavior history in various credit services, building an evaluation method for user preferences for credit service models is a key step in credit business recommendations, which can effectively improve the accuracy of credit business recommendations.

[0231] Based on this, in the embodiment of the present invention, the method for evaluating the credit service model preference (also known as credit willingness) is to construct an integrated linkage evaluation method of a machine learning classification model and a time series model. Among them, the classification model is used to predict the probability of a user using a certain credit service model and the specific subdivision categories under the model; the time series model is used to predict the specific usage indicators of a user under a certain credit service model, such as credit usage limit, usage cycle, usage interest, etc.

[0232] Specifically, in step 4, if Figure 7 As shown in the figure, user data and credit business behavior feedback data are used to build a credit service model preference evaluation model (that is, to integrate the machine learning classification model with the time series model), and a credit service model to credit business preference mapping model is built. Finally, these two models are used to give a credit business willingness evaluation score for a certain user.

[0233] Specifically, for a certain user, the process of evaluating the credit business willingness evaluation score is as follows: Figure 8 As shown, including:

[0234] First, the user data (i.e. the historical behavior data generated by the operator) is integrated with the credit business server feedback data (i.e. the credit business historical behavior data) to perform data preprocessing (data cleaning, data integration, data transformation, data reduction, etc.).

[0235] Among them, data cleaning refers to filtering duplicate data, missing data, and abnormal data; data integration refers to matching users of multiple data sources (credit assessment data, business service data, demand assessment data, etc.) and integrating redundant indicators from different data sources; data transformation refers to normalizing data to facilitate subsequent mining and processing needs; data reduction refers to creating new indicator dimensions through methods such as attribute merging, decision tree induction, and principal component analysis, thereby improving data mining efficiency and reducing model processing costs.

[0236] Then the preprocessed data is input into the classifier model and the time series model for parallel processing:

[0237] The classifier model extracts user behavior statistical features from the preprocessed data to obtain the feature vector CrdModFt, and then classifies it through a pre-trained credit service model classifier (when training the classifier, it is necessary to determine the feature vector of the sample data set and the label data CrdModLb (which can be understood as classification)) to obtain the classification result CrdModRst and the effect evaluation index CrdModRstPrb (such as confidence, etc.) corresponding to each result, so as to facilitate the subsequent fusion of the evaluation results.

[0238] The time series model extracts time series features from the preprocessed data to obtain the time series vector TmSrFt, and then uses the time series model to perform indicator prediction TmSrRst to obtain the prediction results and the corresponding effect evaluation indicators TmSrRstPrb for subsequent fusion of the evaluation results.

[0239] Finally, the evaluation results are effectively fused through the evaluation result fusion processing step to obtain more accurate mode preference evaluation results.

[0240] The core processing logic includes: each classification / prediction result is accompanied by a matching effect evaluation indicator, such as confidence. The method of fusing the evaluation results is to standardize the "effect evaluation indicators" of the two models, and then compare them based on the standardized processing results to obtain the classification / prediction result with a better (more reliable) effect evaluation indicator. The standardized processing method is designed to statistically calculate the distribution of effect evaluation indicators of each classification / prediction result, and then map their respective distributions to a unified normal distribution, so that the effect evaluation indicators of the parallel processing models are comparable.

[0241] Here, the construction of the classifier model includes:

[0242] Step 1: Data acquisition and feature construction;

[0243] Specifically, the communication / consumption / surfing / location data (i.e. user data) generated by the user at the communication operator and the credit service usage behavior data (i.e. credit service historical behavior data) generated at each credit service provider are obtained to construct feature vector data and label data.

[0244] The eigenvector CrdModFt is obtained by:

[0245] CrdModFt = [communication / consumption / surfing / location and other data indicators, business 1 credit behavior data indicators, business 2 credit behavior data indicators, ..., business N credit behavior data indicators]

[0246] The credit behavior data indicators of business i include indicators based on credit ability assessment and indicators based on actual user usage behavior, including but not limited to the indicators shown in Table 4:

[0247]

[0248]

[0249] Table 4

[0250] The label data CrdModLb is obtained by the following method:

[0251] The user's label data on each type of service mode is formed by the statistics of the various business usage records contained in each type of credit service mode. The label data is classified into two categories, one is the label of whether service mode j is used, and the other is the label of the usage preference category under mode j, as shown in Table 5:

[0252]

[0253] Table 5

[0254] Step 2: Classifier model construction;

[0255] like Fig. 9 As shown, for Q service modes, Q sets of classifiers are constructed in parallel. For service modes with subdivided categories, subclassifiers are constructed to evaluate the subdivided mode categories. The methods that can be used by each subclassifier include but are not limited to logistic regression classification model, tree structure classification model, neural network classification model, integrated classification model and other methods. The embodiments of the present invention are not limited to this.

[0256] The user feature vector data CrdModFt and each mode label data CrdModLb are respectively input into the established classifier model to obtain the usage probability of each credit service mode, the category of specific usage indicators under each credit service mode, and the effect evaluation indicators of each category result:

[0257] CrdModRst=[Mod1Rst, Mod2Rst,..., ModMRst]

[0258] CrdModRst=[Mod1RstPrb, Mod2RstPrb,..., ModMRstPrb]

[0259] For example, if mode 1 is an installment purchase mode, including three indicators, namely, installment amount category C, installment cycle category T, and installment interest category I, the output data of mode 1 is (P 1 , C 1 , T 1 , I 1); If mode 2 is a deposit-free mode, including a deposit-free quota category C, the output data of mode 1 is (P 2 , C 2 ); Mode 3 is free of pre-storage and has no subdivision index, so the output data of Mode 3 is P 3 The total output result is:

[0260] CrdModRst=[(P 1 , C 1 , T 1 , I 1 ), (P 2 , C 2 ), (P 3 )...(P M )]

[0261] CrdModRstPrb=[(PC 1 , PT 1 , P.I. 1 ), (PC 2 )...]

[0262] The construction of time series models includes:

[0263] Since users’ preferences for various credit service models vary over time in some indicators, for example, the use of credit installment amounts may be related to the changes in users’ income in different months. Therefore, a prediction model for specific indicators under the credit service model is constructed based on time series, such as Fig.10 shown.

[0264] Time series models are used to predict specific indicators under various service models, such as usage quota prediction, usage cycle prediction, usage interest prediction, etc.

[0265] The input data is the time series of a certain indicator of the user in the credit service mode, a group of variables arranged in chronological order, using TmSrFt = [x t-1 , x t-2 , ..., x t-q ] to indicate.

[0266] Time series processing and prediction are performed through time series models such as trend fitting, ARMA model, ARIMA model, ARCH model and other methods.

[0267] For example, using Fig.11 The process shown in the figure constructs the following ARMA-based model to predict indicators under the credit service model:

[0268]

[0269] Finally, we can obtain the predicted values ​​of specific behavior indicators and evaluation effect indicators of users under a certain credit model. The predicted values ​​of specific behavior indicators are as follows: j , period prediction value T j , Interest usage prediction value I j Etc., evaluation effect indicators such as the confidence level of each prediction value PC j , PT j ,PI j wait.

[0270] The results are as follows:

[0271] TmSrRst=[(C 1 , T 1 , I 1 ), (C 2 ), ...]

[0272] TmSrRstPrb=[PC 1 , PT 1 , P.I. 1 ), (PC 2 )...]

[0273] The evaluation results are effectively fused through the evaluation result fusion processing step to obtain more accurate mode preference evaluation results.

[0274] Specifically, first obtain the effect evaluation index that matches each classification / prediction result, such as the confidence index of each evaluation result, such as PC j , PT j ,PI j wait.

[0275] The "effect evaluation index" of the two models is standardized respectively. The standardization method is designed to count the distribution of the effect evaluation index of each classification / prediction result on the sample, and then map each distribution to a normal distribution with unified parameters, so that the effect evaluation indexes of the parallel processing models are comparable. PCZ is obtained in this way. j , PTZ j ,PIZ j wait.

[0276] Then compare the standardized results of the two models and take the classification / prediction result with better (more reliable) standardized effect evaluation indicators. For example, compare the PCZ values ​​output by the two models and take the C value output by the model with the larger PCZ value. The processing logic of other indicators such as T and I is the same. Finally, CrdDsRst is obtained:

[0277] CrdDsRst=(Mod1FusRst, Mod2FusRst,..., ModMFusRst)

[0278] For example, CrdDsRst=((P 1 , C 1 , T 1 , I 1 ), (P 2 , C 2 ), (P 3 ),…(P M ))

[0279] Among them, the values ​​of C, T, and I are the fusion results after comparison of standardized effect evaluation indicators.

[0280] After obtaining the above results, it is necessary to map the credit service model preference evaluation results to the credit business preferences.

[0281] like Fig.12 As shown, the mapping process includes: first determine whether the credit service model preference assessment results have subdivision categories, such as Figure 2 For international roaming, calls / surfing, mobile phones, etc. in the prepaid mode shown, if yes, use the user credit service mode preference and credit business mode similarity matching model to perform similarity matching between the service mode corresponding to the user preference and the service mode of the credit business, and use the score mapping function to perform score mapping based on the matching result. If there is no subdivision category, use the score mapping function directly to perform score mapping.

[0282] Among them, the first thing to do is to obtain data, that is, the user's preference evaluation index CrdDsRst for the credit service model to which the credit business belongs (the above-mentioned usage probability of the credit model, usage limit prediction results, usage cycle prediction results, usage interest prediction results, etc.).

[0283] For credit service mode j without subdivision (such as pre-deposit-free service category), construct the mode usage probability P j The mapping function to the preference score Scrd_ds = g1(P j ). The probability of using the credit mode P j The mode preference score Scrd_ds is obtained through the score mapping function.

[0284] For credit service models j with subdivided categories (such as deposit-free service categories, credit installment service categories, etc.), similarity measurement methods (such as distance function calculation) are used to calculate the similarity between each credit service model and the credit business. Then, based on the similarity index, model probability P j , through the score mapping function, it is mapped to the user's mode preference score for credit business Scrd_ds.

[0285] Specifically, construct a similarity calculation function Sim between the pattern vector of pattern j and the pattern vector of credit business i ij =f(MV j , BMV i ). For example, the mode vector of credit mode j is MV j =(C j , T j , I j ), the pattern vector BMV of credit business i j =(C i , T i , I i ), similarity measurement based on distance function:

[0286]

[0287] Among them, A is a constant parameter, and fd is the distance mapping function (adjusting the similarity distribution).

[0288] Then, construct the similarity to preference score mapping function Scrd_ds i =g2(P j ,Sim ij ), according to P j and similaritySim ij After processing by g2, the preference score Scrd_ds is obtained.

[0289] After the above similarity matching and score mapping model processing, the preference score of each credit business, namely the willingness score Scrd_ds, is obtained.

[0290] Scrd ds =(Scrd_ds 1 ,Scrd_ds 2 ,...Scrd_ds M )

[0291] In step 5, the demand assessment for user credit services refers to the assessment of specific business-level needs such as purchasing a terminal, renting a car, staying in a hotel, and reading.

[0292] In the embodiment of the present invention, when establishing the customer demand preference label at the business layer, a hybrid feature mining method based on "expert knowledge rules + deep learning recognition" is adopted. The customer's communication behavior is profiled based on the Internet access data of the customer's mobile terminal and home broadband, telephone communication data, etc., and the demand preference is labeled. After calculating the similarity of the label vector with each credit business one by one, the top P (TOP-P) business recommendation list is formed according to the similarity from high to low.

[0293] Among them, the expert recommendation rule model adopts a manually maintained fixed recommendation strategy, and gives recommendation results only by searching for customer interest keywords, daily visited websites, and daily used applications (APP) statistics. This method is simple to implement and has a fast recommendation speed. It is very effective for scenarios with a small number of recommended business items, and can still give recommendation strategies when there is less customer communication behavior data. However, since it is difficult to continuously maintain keywords, website parsing libraries, and APP parsing libraries (such as incomplete keywords and low update frequency), and it is impossible to model the timeline of customer access behavior, the recommendation accuracy is not high and the coverage is incomplete.

[0294] Feature recognition based on deep learning uses machine learning data processing technology to form a denser high-level semantic abstraction by combining low-level features of customer communication behaviors, automatically discover distributed feature representations of data, and solve the problem of manually designing features and rules in expert knowledge-based recommendations. It has wide coverage and high accuracy. However, deep learning recognition (i.e., deep learning model) has the disadvantages of requiring a large amount of customer data to train offline recommendation models, and once the model is established, it is necessary to collect data again to train the model when new recommended credit businesses are added. It is not as easy to adjust as expert rules and cannot adapt to business adjustments and changes in a timely manner. In addition, when mining new customer demand features, it is necessary to collect sufficient customer behavior input so that the model output will be more accurate.

[0295] Therefore, in the embodiment of the present invention, the collaborative linkage method (hybrid recommendation method) of "expert knowledge rules + deep learning" is used for evaluation; that is, on the one hand, "expert knowledge" is used to achieve fast cold start recommendations and recommendations for new business scenarios, and on the other hand, "deep learning" is used to make up for the lack of expert knowledge in extracting customer demand features and insight depth. "Expert knowledge" will be realized by constructing customer demand labels and user matching rules, and "deep learning recognition" will use LDA+LSTM to build a two-layer connection hybrid demand feature recognition model.

[0296] The collaborative recommendation process of "expert knowledge rules + deep learning" is as follows: Fig.13 As shown, it includes: taking the deep learning model as the highest priority, if the data is sufficient and the recommended items meet the filtering rules (such as the confidence is higher than the preset threshold, etc.), the recommendation results of deep learning are adopted; if the deep learning model conditions are not met, the data is input into the expert rule recommendation model, if the expert rules meet the filtering rules, the expert rule model recommendation results are adopted; if the expert rule recommendation model conditions are not met, the random strategy recommendation model is adopted. Through the multi-model collaboration mechanism, it can be ensured that users with different data conditions and evaluation conditions can get the best recommendation results within their evaluable range.

[0297] Among them, the construction of the expert rule recommendation model includes:

[0298] Step 1: Construct customer demand labels and user data matching rules;

[0299] First, collect customer Internet behavior data, terminal data, communication data, etc., as well as the credit service usage behavior data generated by each credit service provider to build demand labels, and build a customer demand assessment model for credit services based on the demand labels.

[0300] First, we build a user demand tag library around the credit business. Demand tags can reflect customers' interests and hobbies, and their needs for certain products and services, as shown in Table 6:

[0301] Tag ID Requirement Tags T1 Mobile terminal T2 read T3 call … …

[0302] Table 6

[0303] Then, a matching rule base with the required labels is constructed based on the acquired data.

[0304] The score of a user on each demand tag is obtained by evaluating the indicators of the user hitting each item in the rule library within a certain period of time. Among them, the indicators can be statistical indicators such as the number of visits and the length of the visit. The search keyword matching indicator is represented by Nkw, the web page access matching indicator is represented by Nweb, and the APP access indicator is represented by Napp. The customer's tag score on tag Ti is calculated by the following method:

[0305] St1 Ti =f1(Nkw Ti , Nweb Ti , Napp Ti )

[0306] Among them, f1 can be obtained by weighted summation, expert scoring, hierarchical analysis method and other methods, which are not limited in the embodiment of the present invention.

[0307] When matching the rule base, you can consider the following aspects:

[0308] (1) Establishing the association between demand tags and customer terminal data

[0309] Specifically, based on customer terminal data, an association library between demand tags and terminal data is established. The intensity of customer demand for services such as purchasing a device is evaluated through information such as the frequency of terminal replacement, terminal brand and model, and terminal functions. Examples of indicators are shown in Table 7:

[0310]

[0311]

[0312] Table 7

[0313] Based on the above indicators, the customer's demand for mobile phone terminals is evaluated through expert scoring method, decision tree algorithm, etc. Then the score contribution of terminal data to label Ti is St2 Ti :

[0314]

[0315] Among them, N i It is a label T i The corresponding number of terminal related indicators.

[0316] (2) Establishing the association between demand tags and customer communication data

[0317] Based on customer communication data, establish a database of associations between demand tags and communication data. Evaluate the demand intensity of communication services through information such as communication behavior data. Examples of indicators are shown in Table 8:

[0318] Tag ID Requirement Tags Terminal indicators Terminal data indicator examples illustrate T3 call <![CDATA[Tele1 T3 ]]> Call duration, number of calls, etc. Debt payment indicators

[0319] Table 8

[0320] Based on the above indicators, the customer's demand for communication services such as calls is evaluated through expert scoring method and decision tree algorithm. Then the score contribution of communication data to label Ti is St3 Ti :

[0321]

[0322] Among them, M i It is the number of communication-related indicators corresponding to the label Ti.

[0323] Step 2: Build a demand label scoring model;

[0324] Specifically, the score St obtained by the customer on the label Ti Ti The calculation is as follows:

[0325] St Ti =g(st1 Ti , St2 Ti , St3 Ti )

[0326] Among them, g can be a mapping function obtained by weighted average method, expert scoring method, hierarchical analysis method (AHP method), decision tree algorithm and other methods. The embodiment of the present application does not limit the specific calculation method.

[0327] For example, St Ti =w 1 *St1 Ti +w 2 *St2 Ti +w 3*St3 Ti

[0328] Among them, w 1 、w 2 、w 3 They are the weights of the influence of Internet behavior score, terminal data score and communication data score on the overall score of the label.

[0329] Step 3: Build the correlation between demand labels and credit business.

[0330] Specifically, after obtaining the scores of each label in the demand label system, it is necessary to map the label scores to the customer's demand intensity for each credit business, so it is necessary to construct a correlation indicator between the demand label and the credit business.

[0331] The correlation index is constructed as follows:

[0332] Based on the distribution of credit scores, a sample of customers with credit business handling records is extracted in a stratified manner to form a demand assessment sample set D3;

[0333] Based on the number of transactions, usage quota, etc., a comprehensive index H of the usage records of customers in each credit business in the sample set D3 is constructed. j ;

[0334] Extract online behavior data, terminal data, communication data, etc. from customers in sample set D3;

[0335] Based on the data obtained in the previous step, calculate St1 of the customers in sample set D3 respectively Ti , St2 Ti , St3 Ti , and based on St Ti =g(St1 Ti , St2 Ti , St3 Ti ) Get the score St of each customer on each tag Ti ;

[0336] For H j Respectively with each label score St Ti For correlation analysis, Pearson correlation coefficient (person), Spearman correlation coefficient (spearman), etc. may be used, and the specific processing method is not limited in the embodiment of the present invention.

[0337] Among them, the higher the correlation index, the greater the impact of the demand label on the demand assessment of the credit business. For example, the "call" label has a high correlation with the "credit without stopping" business, and the "reading" label has a high correlation with the "credit borrowing book" business (0.9). The lower the correlation index, the smaller the impact of the demand label on the demand assessment of the credit business. For example, the "reading" label has a low correlation with the "credit without stopping" business, and the "call" label has a low correlation with the "credit borrowing book" business (0.1).

[0338] After distinguishing the first and second types of business according to the above description, based on label scoring and correlation analysis, the method for evaluating the user's demand score for a certain credit business is as follows:

[0339] User demand for the first type of service is divided into Sbsn_ds Aj The demand for the second type of business is divided into Sbsn_ds Bj Calculation method:

[0340] Sbsn_ds Aj =∑ Ti St Ti *CorA ij ;

[0341]

[0342] The final processing results are as follows:

[0343] For a certain user, the processing results of this step are as follows:

[0344] Sbsn_ds=[(A1, Sbsn_ds1 A1 ), (A2, Sbsn_ds1 A2 ), (B1, Sbsn_ds1 B1 ), (B2, Sbsn_ds1 B2 )]

[0345] The deep learning model of the embodiment of the present invention is a two-layer deep learning credit service recommendation network based on LDA-LSTM, which can also be called an LDA-LSTM classifier (model), which can realize customer preference feature extraction and credit service recommendation. The LDA model can be responsible for feature extraction and topic induction of access content, and the LSTM module can be responsible for discovering the inherent rules in the user's browsing and clicking sequence. The model is updated and iterated according to the new data that is continuously added to achieve better model accuracy, such as Fig.14 shown.

[0346] In actual application, the Internet log system can be used to collect customer request information for accessing web addresses and apps, obtain web and h5 URL information, access time, access sequence, and access frequency, and then use web crawlers to detect the specific content and key fields of these addresses and use these data as training data.

[0347] Based on the output probability, the score mapping is performed. For a certain user, the processing result after this step is as follows:

[0348] Sbsn_ds=[(A1, Sbsn_ds2 A1 ), (A2, Sbsn_ds2 A2 ), (B1, Sbsn_ds2 B1 ), (B2, Sbsn_ds2 B2 )]

[0349] Based on the above demand assessment results, the final result of the business demand assessment can be expressed as:

[0350] Sbsn_ds=[(A1, Sbsn_ds A1 ), (A2, Sbsn_ds A2 ), (B1, Sbsn_ds B1 ), (B2, Sbsn_ds B2 )].

[0351] In step 6, the obtained credit business capability assessment score Scrd_ab, the credit business willingness assessment score Scrd_ds and the credit business demand assessment score Sbsn_ds are fused and calculated to obtain personalized recommendation results.

[0352] Exemplarily, these three scores are weighted to obtain a comprehensive recommendation score for the corresponding credit business:

[0353]

[0354] Srcm Bi =w1*Scrd_ab Bi +w2*Scrd_ds Bi +w3*Sbsn_ds Bi

[0355] Finally, based on the comprehensive recommendation score, the credit businesses are sorted from high to low to obtain the sorting results, thereby obtaining personalized recommendation results for customers.

[0356] In step 7, when the client displays the credit business, the credit business is displayed according to the personalized recommendation sorting result, that is, the credit business is displayed according to the sorting result of step 6; that is, the business with high service level and high demand is ranked in a high priority position, and the business with low service level and low demand is ranked in a low priority position.

[0357] When displaying, the user's service evaluation results (service level / quota and other information) are displayed in the service recommendation list to facilitate the user to determine whether to apply for a certain service.

[0358] In addition, in the embodiment of the present invention, Fig.15 As shown in the figure, a business behavior feedback closed loop and model iteration and tuning mechanism are adopted, that is, the credit business behavior data is fed back to each link of the evaluation model (capability evaluation link, willingness evaluation link, and post-evaluation tuning link) to achieve the effect of continuous iteration and tuning of the evaluation model.

[0359] Among them, the feedback mechanism refers to: credit business feedback. Specifically, the business party server feeds back the credit business behavior data to the credit assessment server, forming a data closed-loop feedback mechanism.

[0360] The feedback data includes three categories, namely credit business performance and default data, credit business usage index data, and credit business recommendation response index data. The feedback data are respectively input into the capability assessment link, willingness assessment link, and post-assessment link for model iteration.

[0361] For example, the feedback data is shown in Table 9:

[0362]

[0363] Table 9

[0364] Model iterative optimization includes: pre-evaluation stage (credit business capability sub-evaluation, credit business willingness sub-evaluation) through the user's input and training of credit business behavior data, the evaluation model parameters can be continuously iterated and optimized to obtain more accurate evaluation results; post-evaluation stage, through the input of user response behavior data to the recommendation results, a post-evaluation processing method is constructed to achieve iterative tuning of business recommendation results.

[0365] Among them, post-evaluation processing methods, such as Fig.16 As shown, including:

[0366] Obtain the response indicators of the business recommendation results (i.e., the indicators shown in Table 9), and then standardize the indicators. Determine the influence weight of each indicator based on machine learning or expert knowledge, and calculate the comprehensive influence index of the feedback indicators on each credit business through weighted summation and other methods according to the influence weight. Then, determine the impact authority (such as impact percentage, etc.) based on expert knowledge and other methods, adjust the comprehensive impact index through the impact authority, and add the adjusted result to the final score of each business to obtain the adjusted score. Finally, sort according to the adjusted score to obtain the final recommendation result.

[0367] From the above description, it can be seen that the solution of the embodiment of the present invention constructs a multi-level evaluation model of "credit ability", "credit service model preference", and "business demand", which effectively combines the evaluation of the applicable scope of user ability, credit service scenarios, and specific business scenario needs; in the evaluation of credit service scenarios, the "classifier + time series" evaluation method is adopted to more accurately evaluate the user's preference for credit business use. Through the multi-level evaluation model of credit ability, credit service model preference, and credit business demand, not only can preference evaluation and recommendation be performed on the granularity of large classification scenarios of credit services, such as deposit-free scenarios and credit installment scenarios, but also the evaluation of specific credit businesses (such as deposit-free car rental and deposit-free cycling), the evaluation of specific segmented business scenarios (such as large-amount installment preference and small-amount installment preference), etc., can be realized, effectively improving the accuracy of credit business evaluation and recommendation.

[0368] At the same time, the optimization method of the credit capability assessment linkage processing steps and the cold start method based on the expert knowledge base in the business needs assessment, that is, building a credit business capability rule base, offline database and system linkage processing steps optimization, can effectively improve the recommendation processing speed.

[0369] For example, assume that user A and user B have the following characteristics:

[0370] User A is a business traveler with high communication consumption, frequent phone calls, high demand for outbound traffic, sufficient funds, preference for postpaid and deposit-free services, high demand for car rentals and hotel accommodations when traveling on business trips;

[0371] User B is a campus student who has high capital turnover needs, is accustomed to using credit installment purchase services, is interested in new devices and new mobile phones, loves reading, and frequently travels by bicycle.

[0372] When the indifferent recommendation and display results of the relevant technical solution are as follows, different users obtain the same credit business recommendation results, such as Fig.17 shown.

[0373] When the solution of the embodiment of the present invention is adopted, personalized recommendation results of credit services can be obtained according to the needs and preferences of users. That is, according to the credit service evaluation of users A and B and the user needs evaluation, personalized recommendation rankings of A and B are obtained:

[0374] User A (“Credit without downtime”, “Credit car rental”, “Credit hotel stay”, …)

[0375] User B ("credit purchase", "credit reading", "credit riding", ...)

[0376] The displayed information should include the ranking of the recommended services, the name of the recommended services, the level of the recommended services evaluated by the user, etc., so that users can efficiently know their service level in each service, such as Fig.18 shown.

[0377] In order to implement the method of the embodiment of the present invention, the embodiment of the present invention further provides a service recommendation device, which can be set on a server, such as Fig.19 As shown, the device comprises:

[0378] An acquisition unit 191 is configured to acquire a credit parameter of the first user; the credit parameter represents the credit ability of the first user;

[0379] A first determining unit 192, configured to determine a credit service that the first user can use by using the acquired credit parameter and the identity information of the first user;

[0380] The second determination unit 193 is used to determine the credit service mode preference of the first user by using the first data and the second data of the first user; and determine the credit service demand of the first user by using the first data and the second data; the first data represents the credit service historical behavior of the first user; the second data represents the historical behavior of the first user in the communication operator;

[0381] The third determining unit 194 is configured to recommend a credit service to the first user by using the determined credit services available to the first user, the credit service mode preference, and the credit service requirements.

[0382] In one embodiment, the first determining unit 192 is specifically configured to:

[0383] For each of at least two types of credit services in the database, determine from the corresponding type of credit services a service that matches the acquired credit parameters and a corresponding service level, or determine from the corresponding type of credit services a service that matches the identity information of the first user and a corresponding service level.

[0384] In one embodiment, the second determining unit 193 is specifically configured to:

[0385] Performing a first feature extraction on the first data and the second data to obtain a first extraction result, and performing a second feature extraction on the first data and the second data to obtain a second extraction result; the first extraction result represents the behavior of the first user; and the second extraction result represents the time series of the first user;

[0386] Using the first extraction result in combination with the first model, evaluating the credit service mode preference of the first user to obtain a first credit service mode preference result;

[0387] Using the second extraction result and a second model, evaluating the credit service mode preference of the first user to obtain a second credit service mode preference result;

[0388] The credit service mode preference of the first user is determined by using the first credit service mode preference result and the second credit service mode preference result.

[0389] In one embodiment, the determining the credit service mode preference of the first user by using the first credit service mode preference result and the second credit service mode preference result includes:

[0390] The second determining unit 193 normalizes the first credit service mode preference result to obtain a first processing result; and normalizes the second credit service mode preference result to obtain a second processing result;

[0391] The second determining unit 193 determines the credit service mode preference of the first user by using the first processing result and the second processing result.

[0392] In one embodiment, the second determining unit 193 is specifically configured to:

[0393] Extracting features from the first data and the second data to obtain user demand features;

[0394] When the obtained user demand characteristics meet the first condition, the obtained user demand characteristics are input into a deep learning model to obtain the credit business demand of the first user; the deep learning model is established based on the LDA-LSTM algorithm using training samples;

[0395] In the case where the obtained user demand feature meets the second condition, the obtained user demand feature is input into the expert rule recommendation model to obtain the credit business demand of the first user; the expert rule recommendation model is constructed by using training samples to construct demand labels and is established based on the constructed demand labels;

[0396] In the case where the obtained user demand characteristics do not meet the first condition and the second condition, the obtained user demand characteristics are input into a random strategy recommendation model to obtain the credit business demand of the first user; the random strategy recommendation model is obtained based on the ranking result of the recommendation probability;

[0397] The first condition is different from the second condition.

[0398] In one embodiment, the second determining unit 193 is further configured to:

[0399] Determining that the number of the obtained user demand features satisfies a third condition; the third condition indicates that the number of the obtained user demand features is sufficient;

[0400] The obtained user demand features are matched with the recommendation item filtering rules corresponding to the deep learning model, and after a successful match, it is determined that the obtained user demand features meet the first condition.

[0401] Among them, the second determination unit 193 matches the obtained user demand features with the recommendation item filtering rules corresponding to the deep learning model. If the matching fails, the obtained user demand features are matched with the recommendation item filtering rules corresponding to the expert rule recommendation model. If the matching succeeds, it is determined that the obtained user demand features meet the second condition.

[0402] In one embodiment, the second determining unit 193 is further configured to:

[0403] Determine that the number of the obtained user demand features does not satisfy the third condition; match the obtained user demand features with the recommendation item filtering rules corresponding to the expert rule recommendation model, and after a successful match, determine that the obtained user demand features meet the second condition.

[0404] The second determining unit 193 matches the obtained user demand feature with the recommended item filtering rule corresponding to the expert rule recommendation model, and if the matching fails, determines that the obtained user demand feature does not meet the first condition and the second condition.

[0405] In one embodiment, the third determining unit 194 is specifically configured to:

[0406] Merging the scores corresponding to each credit service among the credit services available to the first user, the scores corresponding to each credit service among the credit services corresponding to the credit service mode preference, and the scores corresponding to each credit service in the credit service demand to obtain a fusion result;

[0407] A credit service recommendation is made to the first user based on the fusion result.

[0408] The second determination unit 193 performs similarity matching between the credit service model preference and the service model of the credit services in the database, and determines the score corresponding to each credit service in the credit services corresponding to the credit service model preference based on the similarity matching result.

[0409] In one embodiment, the device may further include: a feedback processing unit, configured to:

[0410] Acquiring user data generated by the first user for a first credit service; the first credit service is a credit service selected by the first user after a credit service recommendation is made to the first user;

[0411] Use the generated user data to update the corresponding model.

[0412] In one embodiment, the device may further include:

[0413] The recommendation unit is used to send credit service recommendation information to the client device corresponding to the first user; the service recommendation information includes credit service and corresponding grade information.

[0414] In actual application, the acquisition unit 191 and the feedback processing unit can be implemented by a processor in the service recommendation device in combination with a communication interface; the first determination unit 192, the second determination unit 193, and the third determination unit 194 can be implemented by a processor in the service recommendation device.

[0415] In order to implement the method of the embodiment of the present invention, the embodiment of the present invention further provides a service recommendation device, which is arranged on a client device, such as Fig. 20 As shown, the device comprises:

[0416] The transceiver unit 201 is used to receive the credit service recommendation information sent by the server; the service recommendation information includes the credit service and the corresponding level information;

[0417] The processing unit 202 is used to present a credit service recommendation based on the received credit service recommendation information; the presented content includes the credit service and the corresponding grade information.

[0418] In one embodiment, the transceiver unit 201 is further configured to send the identity information of the first user to the server.

[0419] In actual application, the transceiver unit 201 may be implemented by a communication interface in the service recommendation device; and the processing unit 202 may be implemented by a processor in the service recommendation device.

[0420] It should be noted that: the business recommendation device provided in the above embodiment only uses the division of the above program modules as an example when making business recommendations. In actual applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device is divided into different program modules to complete all or part of the above-described processing. In addition, the business recommendation device provided in the above embodiment and the business recommendation method embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0421] Based on the hardware implementation of the above program modules, and in order to implement the method of the terminal side of the embodiment of the present invention, the embodiment of the present invention also provides a server, such as Fig.21 As shown, the server 210 includes:

[0422] The first communication interface 211 is used to exchange information with other devices to make service recommendations;

[0423] The first processor 212 is connected to the first communication interface 211 to implement information exchange with other devices and is used to execute the method provided by one or more technical solutions on the server side when running a computer program. The computer program is stored in the first memory 213.

[0424] Specifically, the first processor 212 is configured to:

[0425] Acquiring a credit parameter of the first user; the credit parameter represents the credit ability of the first user;

[0426] Determining a credit service that the first user can use by using the acquired credit parameter and the identity information of the first user;

[0427] Determine the credit service mode preference of the first user by using the first data and the second data of the first user; and determine the credit service demand of the first user by using the first data and the second data; the first data represents the credit service historical behavior of the first user; the second data represents the historical behavior of the first user in the communication operator;

[0428] The determined credit services available to the first user, the credit service model preference, and the credit service needs are used to recommend credit services to the first user.

[0429] In one embodiment, the first processor 212 is specifically configured to:

[0430] For each of at least two types of credit services in the database, determine from the corresponding type of credit services a service that matches the acquired credit parameters and a corresponding service level, or determine from the corresponding type of credit services a service that matches the identity information of the first user and a corresponding service level.

[0431] In one embodiment, the first processor 212 is specifically configured to:

[0432] Performing a first feature extraction on the first data and the second data to obtain a first extraction result, and performing a second feature extraction on the first data and the second data to obtain a second extraction result; the first extraction result represents the behavior of the first user; and the second extraction result represents the time series of the first user;

[0433] Using the first extraction result in combination with the first model, evaluating the credit service mode preference of the first user to obtain a first credit service mode preference result;

[0434] Using the second extraction result and a second model, evaluating the credit service mode preference of the first user to obtain a second credit service mode preference result;

[0435] The credit service mode preference of the first user is determined by using the first credit service mode preference result and the second credit service mode preference result.

[0436] In one embodiment, the determining the credit service mode preference of the first user by using the first credit service mode preference result and the second credit service mode preference result includes:

[0437] The first processor 212 normalizes the first credit service mode preference result to obtain a first processing result; and normalizes the second credit service mode preference result to obtain a second processing result;

[0438] The first processor 212 determines the credit service mode preference of the first user by using the first processing result and the second processing result.

[0439] In one embodiment, the first processor 212 is specifically configured to:

[0440] Extracting features from the first data and the second data to obtain user demand features;

[0441] When the obtained user demand characteristics meet the first condition, the obtained user demand characteristics are input into a deep learning model to obtain the credit business demand of the first user; the deep learning model is established based on the LDA-LSTM algorithm using training samples;

[0442] In the case where the obtained user demand feature meets the second condition, the obtained user demand feature is input into the expert rule recommendation model to obtain the credit business demand of the first user; the expert rule recommendation model is constructed by using training samples to construct demand labels and is established based on the constructed demand labels;

[0443] In the case where the obtained user demand characteristics do not meet the first condition and the second condition, the obtained user demand characteristics are input into a random strategy recommendation model to obtain the credit business demand of the first user; the random strategy recommendation model is obtained based on the ranking result of the recommendation probability;

[0444] The first condition is different from the second condition.

[0445] In one embodiment, the first processor 212 is further configured to:

[0446] Determining that the number of the obtained user demand features satisfies a third condition; the third condition indicates that the number of the obtained user demand features is sufficient;

[0447] The obtained user demand features are matched with the recommendation item filtering rules corresponding to the deep learning model, and after a successful match, it is determined that the obtained user demand features meet the first condition.

[0448] Among them, the first processor 212 matches the obtained user demand features with the recommended item filtering rules corresponding to the deep learning model. If the matching fails, the obtained user demand features are matched with the recommended item filtering rules corresponding to the expert rule recommendation model. If the matching succeeds, it is determined that the obtained user demand features meet the second condition.

[0449] In one embodiment, the first processor 212 is further configured to:

[0450] Determine that the number of the obtained user demand features does not satisfy the third condition; match the obtained user demand features with the recommendation item filtering rules corresponding to the expert rule recommendation model, and after a successful match, determine that the obtained user demand features meet the second condition.

[0451] The first processor 212 matches the obtained user demand features with the recommended item filtering rules corresponding to the expert rule recommendation model, and if the matching fails, determines that the obtained user demand features do not meet the first condition and the second condition.

[0452] In one embodiment, the first processor 212 is specifically configured to:

[0453] Merging the scores corresponding to each credit service among the credit services available to the first user, the scores corresponding to each credit service among the credit services corresponding to the credit service mode preference, and the scores corresponding to each credit service in the credit service demand to obtain a fusion result;

[0454] A credit service recommendation is made to the first user based on the fusion result.

[0455] The first processor 212 performs similarity matching between the credit service model preference and the service model of the credit services in the database, and determines the score corresponding to each credit service in the credit services corresponding to the credit service model preference based on the similarity matching result.

[0456] In one embodiment, the first processor 212 is further configured to:

[0457] Acquiring user data generated by the first user for a first credit service; the first credit service is a credit service selected by the first user after a credit service recommendation is made to the first user;

[0458] Use the generated user data to update the corresponding model.

[0459] It should be noted that the specific processing process of the first processor 212 is detailed in the method embodiment and will not be repeated here.

[0460] Of course, in actual application, the various components in the server 210 are coupled together through the bus system 214. It is understandable that the bus system 214 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 214 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Fig.21 Various buses are labeled as bus system 214 .

[0461] The first memory 213 in the embodiment of the present invention is used to store various types of data to support the operation of the server 210. Examples of such data include: any computer program used to operate on the server 210.

[0462] The method disclosed in the above embodiment of the present invention can be applied to the first processor 212, or implemented by the first processor 212. The first processor 212 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the first processor 212. The above first processor 212 may be a general processor, a digital signal processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The first processor 212 can implement or execute the methods, steps and logic block diagrams disclosed in the embodiment of the present invention. The general processor may be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as a hardware decoding processor to execute, or a combination of hardware and software modules in the decoding processor to execute. The software module can be located in a storage medium, which is located in the first memory 213. The first processor 212 reads the information in the first memory 213 and completes the steps of the above method in combination with its hardware.

[0463] In an exemplary embodiment, the server 210 may be implemented by one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to execute the aforementioned method.

[0464] Based on the hardware implementation of the above program modules, and in order to implement the method on the client device side of the embodiment of the present invention, the embodiment of the present invention also provides a client device, such as Fig. 22 As shown, the client device 220 includes:

[0465] The second communication interface 221 is capable of exchanging information with the server;

[0466] The second processor 222 is connected to the second communication interface 221 to implement information interaction with the terminal and is used to execute the method provided by one or more technical solutions on the client device side when running a computer program. The computer program is stored in the second memory 223.

[0467] Specifically, the second communication interface 221 is used to receive credit service recommendation information sent by the server; the service recommendation information includes credit services and corresponding level information;

[0468] The second processor 222 is used to present a credit service recommendation based on the received credit service recommendation information; the presented content includes the credit service and the corresponding grade information.

[0469] In one embodiment, the second communication interface 221 is further used to send the identity information of the first user to the server.

[0470] Of course, in actual application, the various components in the client device 220 are coupled together through the bus system 224. It can be understood that the bus system 224 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 224 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Fig. 22 Various buses are labeled as bus system 224 .

[0471] The second memory 223 in the embodiment of the present invention is used to store various types of data to support the operation of the client device 220. Examples of such data include: any computer program used to operate on the client device 220.

[0472] The method disclosed in the above embodiment of the present invention can be applied to the second processor 222, or implemented by the second processor 222. The second processor 222 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the second processor 222. The above second processor 222 may be a general processor, DSP, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The second processor 222 can implement or execute the methods, steps and logic block diagrams disclosed in the embodiment of the present invention. The general processor may be a microprocessor or any conventional processor, etc. In combination with the steps of the method disclosed in the embodiment of the present invention, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in the second memory 223. The second processor 222 reads the information in the second memory 223 and completes the steps of the above method in combination with its hardware.

[0473] In an exemplary embodiment, the client device 220 may be implemented by one or more ASICs, DSPs, PLDs, CPLDs, FPGAs, general purpose processors, controllers, MCUs, Microprocessors, or other electronic components to perform the aforementioned methods.

[0474] It can be understood that the memory (such as the first memory 213 and the second memory 223) of the embodiment of the present invention can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM, SyncLink Dynamic Random Access Memory), and direct RAM bus random access memory (DRRAM, Direct Rambus Random Access Memory).The memories described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memories.

[0475] In an exemplary embodiment, the embodiment of the present invention further provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, for example, including a first memory 213 storing a computer program, and the computer program can be executed by the first processor 212 of the server 210 to complete the steps of the aforementioned server-side method. For another example, a second memory 223 storing a computer program can be executed by the second processor 222 of the client device 220 to complete the steps of the aforementioned terminal-side method. The computer-readable storage medium can be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disk, or CD-ROM.

[0476] It should be noted that: "first", "second", etc. are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0477] In addition, the technical solutions described in the embodiments of the present invention can be arbitrarily combined without conflict.

[0478] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention.

Claims

1. A business recommendation method, characterized in that: Applicable to servers, including: Acquiring a credit parameter of the first user; the credit parameter represents the credit ability of the first user; Determining a credit service that the first user can use by using the acquired credit parameter and the identity information of the first user; Determine the credit service mode preference of the first user by using the first data and the second data of the first user; and determine the credit service demand of the first user by using the first data and the second data; the first data represents the credit service historical behavior of the first user; the second data represents the historical behavior of the first user in the communication operator; Using the determined credit services available to the first user, the credit service model preference, and the credit service needs, a credit service recommendation is made to the first user; wherein: The using the first data and the second data of the first user to determine the credit service mode preference of the first user includes: Performing a first feature extraction on the first data and the second data to obtain a first extraction result, and performing a second feature extraction on the first data and the second data to obtain a second extraction result; the first extraction result represents the behavior of the first user; and the second extraction result represents the time series of the first user; Using the first extraction result in combination with the first model, evaluating the credit service mode preference of the first user to obtain a first credit service mode preference result; Using the second extraction result and a second model, evaluating the credit service mode preference of the first user to obtain a second credit service mode preference result; Determine the credit service mode preference of the first user by using the first credit service mode preference result and the second credit service mode preference result; The using the first data and the second data to determine the credit service demand of the first user includes: Extracting features from the first data and the second data to obtain user demand features; When the obtained user demand characteristics meet the first condition, the obtained user demand characteristics are input into a deep learning model to obtain the credit business demand of the first user; the deep learning model is established based on a latent Dirichlet allocation-long short-term memory network LDA-LSTM algorithm using training samples; In the case where the obtained user demand feature meets the second condition, the obtained user demand feature is input into the expert rule recommendation model to obtain the credit business demand of the first user; the expert rule recommendation model is constructed by using training samples to construct demand labels and is established based on the constructed demand labels; The first condition is different from the second condition.

2. The method according to claim 1, characterized in that The using the acquired credit parameter and the identity information of the first user to determine the credit service available to the first user includes: For each of at least two types of credit services in the database, determine from the corresponding type of credit services a service that matches the acquired credit parameters and a corresponding service level, or determine from the corresponding type of credit services a service that matches the identity information of the first user and a corresponding service level.

3. The method according to claim 1, characterized in that The determining the credit service mode preference of the first user by using the first credit service mode preference result and the second credit service mode preference result includes: Normalizing the first credit service model preference result to obtain a first processing result; and normalizing the second credit service model preference result to obtain a second processing result; The credit service mode preference of the first user is determined by using the first processing result and the second processing result.

4. The method according to claim 1, characterized in that: When the credit business demand of the first user obtained by using the deep learning model does not match the recommendation item filtering rule corresponding to the deep learning model, the obtained user demand characteristics are input into the expert rule recommendation model to obtain the credit business demand of the first user.

5. The method according to claim 1, characterized in that When the credit business demand of the first user obtained by using the expert rule recommendation model does not match the recommendation item filtering rule corresponding to the expert rule recommendation model, the obtained user demand characteristics are input into the random strategy recommendation model to obtain the credit business demand of the first user; the random strategy recommendation model is obtained based on the ranking result of the recommendation probability.

6. The method according to claim 1, characterized in that Recommending a credit service to the first user by using the determined credit services available to the first user, the credit service model preference, and the credit service needs, including: Merging the scores corresponding to each credit service among the credit services available to the first user, the scores corresponding to each credit service among the credit services corresponding to the credit service mode preference, and the scores corresponding to each credit service in the credit service demand to obtain a fusion result; A credit service recommendation is made to the first user based on the fusion result.

7. The method according to claim 6, characterized in that The credit service model preference is matched with the service model of the credit business in the database for similarity, and based on the similarity matching result, the score corresponding to each credit business in the credit business corresponding to the credit service model preference is determined.

8. The method according to any one of claims 1 to 7, characterized in that: When recommending a credit service to the first user, the method includes: Sending credit service recommendation information to the client device corresponding to the first user; the credit service recommendation information includes credit service and corresponding grade information.

9. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: Acquiring user data generated by the first user for a first credit service; the first credit service is a credit service selected by the first user after a credit service recommendation is made to the first user; Use the generated user data to update the corresponding model.

10. A business recommendation method, characterized in that: Applied to a client device, the client device corresponding to a first user, comprising: receiving credit service recommendation information sent by a server; the credit service recommendation information includes credit service and corresponding grade information; the credit service recommendation information includes information sent by the server to the client device corresponding to the first user when executing the method of claim 8; Based on the received credit service recommendation information, a credit service recommendation is presented; the presented content includes the credit service and the corresponding grade information.

11. The method according to claim 10, characterized in that The method further comprises: Send the identity information of the first user to the server.

12. A business recommendation device, characterized in that: include: An acquisition unit, configured to acquire a credit parameter of the first user; The credit parameter represents the credit ability of the first user; A first determining unit, configured to determine a credit service that the first user can use by using the acquired credit parameter and the identity information of the first user; a second determining unit, configured to determine the credit service mode preference of the first user by using the first data and the second data of the first user; and using the first data and the second data to determine the credit service demand of the first user; The first data represents the credit business history behavior of the first user; the second data represents the history behavior of the first user in the communication operator; The third determining unit is used to recommend a credit service to the first user by using the determined credit services, credit service model preferences, and credit service requirements that the first user can use; wherein: The second determination unit is specifically used to perform a first feature extraction on the first data and the second data to obtain a first extraction result, and perform a second feature extraction on the first data and the second data to obtain a second extraction result; the first extraction result represents the behavior of the first user; the second extraction result represents the time series of the first user; the first extraction result is used in combination with the first model to evaluate the credit service mode preference of the first user to obtain a first credit service mode preference result; the second extraction result is used in combination with the second model to evaluate the credit service mode preference of the first user to obtain a second credit service mode preference result; the first credit service mode preference result and the second credit service mode preference result are used to determine the credit of the first user. Use service mode preference; and extract features from the first data and the second data to obtain user demand features; when the obtained user demand features meet the first condition, input the obtained user demand features into the deep learning model to obtain the credit business needs of the first user; the deep learning model is established based on the latent Dirichlet distribution-long short-term memory network LDA-LSTM algorithm using training samples; when the obtained user demand features meet the second condition, input the obtained user demand features into the expert rule recommendation model to obtain the credit business needs of the first user; the expert rule recommendation model is established by constructing demand labels using training samples based on the constructed demand labels; wherein the first condition is different from the second condition.

13. A business recommendation device, characterized in that: Set on a client device, the client device corresponds to the first user, including: a transceiver unit, configured to receive credit service recommendation information sent by a server; the credit service recommendation information includes credit service and corresponding grade information; the credit service recommendation information includes information sent by the server to a client device corresponding to the first user when executing the method of claim 8; The processing unit is used to present a credit service recommendation based on the received credit service recommendation information; the presented content includes the credit service and the corresponding grade information.

14. A server, characterized in that: include: a first processor and a first memory for storing a computer program that can be executed on the processor, wherein The first processor is configured to execute the steps of the method according to any one of claims 1 to 9 when running the computer program.

15. A client device, characterized in that: include: a second processor and a second memory for storing a computer program that can be executed on the processor, wherein The second processor is used to execute the steps of the method according to claim 10 or 11 when running the computer program.

16. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 9, or implements the steps of the method according to claim 10 or 11.

Images