Precision recommendation method, device, equipment and medium based on multivariate data fusion

Abstract

The present application proposes a precise recommendation method, device, electronic device and storage medium based on multivariate data fusion. The precise recommendation method based on multivariate data fusion includes: collecting customer data of target customers within a preset time period, and the customer data includes structured data and unstructured data; constructing a first customer feature of the target customer based on the structured data; constructing a second customer feature of the target customer based on the unstructured data; inputting the first customer feature and the second customer feature into the trained precise matching model to obtain the target behavior matching rate of the target customer; comparing the target behavior matching rate with the preset threshold to determine whether the target customer is a customer to be matched, and if the target customer is a customer to be matched, the target customer is pushed to the staff to obtain the target behavior result. The present application can comprehensively consider different types of data to improve the accuracy of precise recommendations.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a precise recommendation method, device, electronic device and storage medium based on multivariate data fusion. Background Art

[0002] The purpose of accurate recommendation is to identify customers who are highly matched with a specific behavior or a specific product, and then push the highly matched customers to relevant staff. It is widely used in marketing, work task allocation and other scenarios. The premise of accurate recommendation is to have available user information data, and then analyze the user information data through rules or models to determine whether the user is suitable for a specific behavior or a specific product.

[0003] At present, in the marketing process of the financial marketing industry, structured data such as basic customer information and click behavior are usually used to mine potential customers to achieve accurate recommendations for marketing customers. However, this method ignores the impact of unstructured data such as historical communication texts on accurate recommendations, resulting in repetitive marketing and reduced accuracy in the accurate recommendation process. Summary of the invention

[0004] In view of the above content, it is necessary to propose a precise recommendation method based on multivariate data fusion and related equipment to solve the technical problem of how to improve the accuracy of precise recommendation. The related equipment includes a precise recommendation device based on multivariate data fusion, an electronic device and a storage medium.

[0005] The present application provides an accurate recommendation method based on multivariate data fusion, the method comprising:

[0006] Collecting customer data of target customers within a preset time period, wherein the customer data includes structured data and unstructured data;

[0007] constructing a first customer feature of the target customer based on the structured data;

[0008] constructing a second customer feature of the target customer based on the unstructured data;

[0009] Inputting the first customer feature and the second customer feature into the trained precise matching model to obtain the target behavior matching rate of the target customer;

[0010] The target behavior matching rate is compared with a preset threshold to determine whether the target customer is a customer to be matched. If the target customer is a customer to be matched, the target customer is pushed to the staff to obtain the target behavior result.

[0011] In some embodiments, collecting customer data of target customers within a preset time period includes:

[0012] Collecting structured data of target customers within a preset time period, wherein the structured data at least includes basic information of the target customers and behavior data on a preset system path, and the target customer is any one of all customers;

[0013] Collecting unstructured data of the target customer within a preset time period, wherein the unstructured data at least includes communication texts between the staff and the target customer;

[0014] The structured data and the unstructured data are stored to obtain customer data of the target customer.

[0015] In some embodiments, constructing the first customer feature of the target customer based on the structured data includes:

[0016] Classifying each type of structured data based on a preset value range of the structured data to obtain a classification result, wherein the classification result includes numerical data and categorical data;

[0017] Based on the classification results, all structured data are subjected to structured preprocessing to obtain a standard value of each type of structured data;

[0018] Numerical encoding is performed on the standard numerical values ​​of the categorical data to obtain characteristic values ​​of each categorical data;

[0019] Binning the standard values ​​of the numerical data to obtain characteristic values ​​of each type of numerical data;

[0020] All feature values ​​are arranged in a fixed order to construct the first customer feature of the target customer.

[0021] In some embodiments, the performing structured preprocessing on all structured data based on the classification result to obtain a standard value of each type of structured data includes:

[0022] Randomly select a type of structured data as the target structured data;

[0023] If the classification result of the target structured data is numerical data, a standard score is calculated according to the actual value of the target structured data; if the standard score is not greater than the preset score, the actual value is used as the standard numerical value of the target structured data; if the standard score is greater than the preset score, the average value of the target structured data in the customer data of all customers is used as the standard numerical value of the target structured data;

[0024] If the classification result of the target structured data is categorical data, determine whether the actual value of the target structured data is within the corresponding preset value range; if so, use the actual value as the standard value of the target structured data; if not, generate a special character as the standard value of the target structured data, and the special character is any character outside the preset value range of the target structured data;

[0025] Traverse all structured data to obtain the standard value of each structured data.

[0026] In some embodiments, the binning of the standard values ​​of the numerical data to obtain the characteristic values ​​of each type of numerical data includes:

[0027] Obtaining a preset value range of target numerical data, wherein the target numerical data is any one of all numerical data;

[0028] Divide the preset value range into a preset number of sub-ranges according to a binning function, and assign different integer data as an index ID to each sub-range;

[0029] Acquire a sub-range corresponding to the standard value of the target numerical data as a target sub-range, and use the index ID of the target sub-range as a characteristic value of the target numerical data;

[0030] Traverse all numerical data to obtain the characteristic value of each numerical data.

[0031] In some embodiments, constructing the second customer feature of the target customer based on the unstructured data includes:

[0032] Performing unstructured preprocessing on the unstructured data to obtain standard sequence data, wherein the unstructured preprocessing includes serialization processing and removing preset characters;

[0033] Splitting the standard sequence data according to a preset rule to obtain a plurality of standard subsequences;

[0034] All standard subsequences are sequentially input into the trained intention recognition model to obtain the intention recognition result of each standard subsequence, and the intention recognition results of all standard subsequences are arranged according to the order of the standard subsequences in the standard sequence data to obtain an intention sequence, wherein the intention recognition result is the customer's demand degree level for the target behavior;

[0035] A second customer profile of the target customer is constructed based on the intention sequence.

[0036] In some embodiments, constructing the second customer feature of the target customer based on the intention sequence includes:

[0037] Acquire the number of types of different intent recognition results in the intent sequence as a first statistical feature;

[0038] Obtaining the intention recognition result that appears most frequently in the intention sequence as a second statistical feature;

[0039] Taking the last intention recognition result in the intention sequence as a time feature;

[0040] A preset number of intention recognition results at the beginning of the intention sequence are intercepted as a starting sequence, and a preset number of intention recognition results at the end of the intention sequence are intercepted as an end sequence;

[0041] Calculating the difference between the proportion of each intention recognition result in the starting sequence and the ending sequence to obtain the trend characteristics of each intention recognition result;

[0042] The first statistical feature, the second statistical feature, the time feature and the trend feature of each intention recognition result are arranged in a fixed order to obtain the second customer feature of the target customer.

[0043] In some embodiments, inputting the first customer feature and the second customer feature into a trained accurate matching model to obtain a target behavior matching rate of the target customer includes:

[0044] Filling zeros at the ends of the first customer feature and the second customer feature to obtain the first customer feature and the second customer feature of a preset size;

[0045] The first customer feature and the second customer feature are fused according to the fusion coefficient to obtain a fusion vector, and the fusion vector satisfies the relationship:

[0046] H=ch1+(1-c)h2

[0047] Wherein, h1 and h2 are the first customer feature and the first customer feature of the preset size respectively, c is the fusion coefficient, the value range of the fusion coefficient is [0,1], and H is the fusion vector;

[0048] The fusion vector is input into the trained precise matching model to obtain the target behavior matching rate of the target customer.

[0049] In some embodiments, the method further includes: updating the precise matching model based on the target behavior result and the target behavior matching rate, specifically including:

[0050] If the target behavior result is successful, the fusion coefficient is adjusted multiple times according to a preset rule, and the first customer feature and the second customer feature are continuously input into the trained precise matching model to obtain target behavior matching rates of multiple adjusted fusion coefficients; the fusion coefficient corresponding to the maximum value of the target behavior matching rate is selected as the new fusion coefficient to complete the update of the precise matching model;

[0051] If the target behavior result is a failure, the label data of the target customer is recorded as 0, and the fusion vector and label data of the target customer are used as a set of new training samples to train the precise matching model, thereby completing the update of the precise matching model;

[0052] The updated precise matching model is used for the next precise recommendation.

[0053] The embodiment of the present application also provides a precise recommendation device based on multivariate data fusion, the device comprising:

[0054] A collection unit, used to collect customer data of target customers within a preset time period, wherein the customer data includes structured data and unstructured data;

[0055] A first construction unit, configured to construct a first customer feature of the target customer based on the structured data;

[0056] A second construction unit, configured to construct a second customer feature of the target customer based on the unstructured data;

[0057] An acquisition unit, configured to input the first customer feature and the second customer feature into a trained accurate matching model to acquire a target behavior matching rate of the target customer;

[0058] The push unit is used to compare the target behavior matching rate with a preset threshold to determine whether the target customer is a customer to be matched. If the target customer is a customer to be matched, the target customer is pushed to the staff to obtain the target behavior result.

[0059] The present application also provides an electronic device, the electronic device comprising:

[0060] A memory storing at least one instruction;

[0061] A processor executes the instructions stored in the memory to implement the precise recommendation method based on multivariate data fusion.

[0062] An embodiment of the present application also provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is executed by a processor in an electronic device to implement the precise recommendation method based on multivariate data fusion.

[0063] In summary, this application comprehensively considers the structured data and unstructured data that affect the accuracy of precise recommendations, and designs different feature extraction methods for different types of data to obtain the first customer feature and the second customer feature. The precise matching model is trained by combining the first customer feature and the second customer feature to realize the screening of all customers and obtain the customers to be matched. At the same time, the precise matching model is updated according to the target behavior results of the customers to be matched, thereby improving the accuracy and timeliness of precise recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a flow chart of a preferred embodiment of the precise recommendation method based on multivariate data fusion involved in this application.

[0065] Figure 2 It is a functional module diagram of a preferred embodiment of the precise recommendation device based on multivariate data fusion involved in this application.

[0066] Figure 3 It is a structural schematic diagram of an electronic device of a preferred embodiment of the precise recommendation method based on multivariate data fusion involved in the present application. DETAILED DESCRIPTION

[0067] In order to more clearly understand the purpose, features and advantages of the present application, the present application is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other. In the following description, many specific details are set forth to facilitate a full understanding of the present application, and the embodiments described are only a part of the embodiments of the present application, rather than all of the embodiments.

[0068] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0069] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art to which the present application belongs. The terms used herein in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0070] The embodiment of the present application provides an accurate recommendation method based on multivariate data fusion, which can be applied to one or more electronic devices. The electronic device is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0071] An electronic device can be any electronic product that can interact with a customer, such as a personal computer, a tablet computer, a smart phone, a personal digital assistant (PDA), a game console, an interactive network television (IPTV), a smart wearable device, etc.

[0072] The electronic device may also include a network device and / or a client device, wherein the network device includes, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud consisting of a large number of hosts or network servers based on cloud computing.

[0073] The network where the electronic device is located includes but is not limited to the Internet, wide area network, metropolitan area network, local area network, virtual private network (VPN), etc.

[0074] like Figure 1 As shown, it is a flow chart of a preferred embodiment of the precise recommendation method based on multivariate data fusion of the present application. According to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted. The precise recommendation method based on multivariate data fusion provided in the embodiment of the present application can be applied to any scenario of precise recommendation of personnel for target behavior, and the method can be applied to products in these scenarios, for example, precise marketing recommendations, precise services, precise allocation of work tasks, etc.

[0075] S10, collecting customer data of target customers within a preset time period, wherein the customer data includes structured data and unstructured data.

[0076] In an optional embodiment, collecting customer data of target customers within a preset time period includes:

[0077] Collecting structured data of target customers within a preset time period, wherein the structured data at least includes basic information of the target customers and behavior data on a preset system path, and the target customer is any one of all customers;

[0078] Collecting unstructured data of the target customer within a preset time period, wherein the unstructured data at least includes communication texts between the staff and the target customer;

[0079] The structured data and the unstructured data are stored to obtain customer data of the target customer.

[0080] Among them, the preset time period is one week, and the preset system path at least includes a software system including a pre-set APP, web page, and public account.

[0081] In this optional embodiment, the basic information of the target customer includes but is not limited to age, gender, occupation and the city where the customer is located; the behavioral data on the preset system path includes but is not limited to browsing time and number of clicks.

[0082] For example, in a precision marketing recommendation scenario, assuming that the target behavior is a marketing behavior of a financial product in the financial marketing industry, the preset system path is a software system including financial management apps, financial management web pages, and financial management public accounts related to the financial management products. By collecting all structured data and unstructured data of the target customer within a week, the customer data of the target customer can be obtained.

[0083] In this way, all the structured and unstructured data collected can more accurately reflect the customer's characteristics from different aspects, ensure the accuracy of subsequent precise recommendations, and provide a data basis for achieving precise recommendations.

[0084] S11, constructing a first customer feature of the target customer based on the structured data.

[0085] In an optional embodiment, constructing the first customer feature of the target customer based on the structured data includes:

[0086] Classifying each type of structured data based on a preset value range of the structured data to obtain a classification result, wherein the classification result includes numerical data and categorical data;

[0087] Based on the classification results, all structured data are subjected to structured preprocessing to obtain a standard value of each type of structured data;

[0088] Numerical encoding is performed on the standard numerical values ​​of the categorical data to obtain characteristic values ​​of each categorical data;

[0089] Binning the standard values ​​of the numerical data to obtain characteristic values ​​of each type of numerical data;

[0090] All feature values ​​are arranged in a fixed order to construct the first customer feature of the target customer.

[0091] In an optional embodiment, classifying each type of structured data based on a preset value range of the structured data to obtain a classification result includes:

[0092] If the preset value range of the structured data is continuous, the structured data is numerical data;

[0093] If the preset value range of the structured data is discrete, the structured data is categorical data.

[0094] The preset value range corresponds to the structured data one by one.

[0095] For example, the preset value range of the city where the customer is located in the basic information of the target customer includes the names of all cities, so the city where the customer is located is categorical data; the preset value range of the browsing time in the behavioral data on the preset system path is continuous, so the browsing time is numerical data.

[0096] In an optional embodiment, the specific methods of structured preprocessing of numerical data and categorical data are different, and the structured preprocessing of all structured data based on the classification result to obtain a standard value of each type of structured data includes:

[0097] Randomly select a type of structured data as the target structured data;

[0098] If the classification result of the target structured data is numerical data, a standard score is calculated according to the actual value of the target structured data; if the standard score is not greater than the preset score, the actual value is used as the standard numerical value of the target structured data; if the standard score is greater than the preset score, the average value of the target structured data in the customer data of all customers is used as the standard numerical value of the target structured data;

[0099] If the classification result of the target structured data is categorical data, determine whether the actual value of the target structured data is within the corresponding preset value range; if so, use the actual value as the standard value of the target structured data; if not, generate a special character as the standard value of the target structured data, and the special character is any character outside the preset value range of the target structured data;

[0100] Traverse all structured data to obtain the standard value of each structured data.

[0101] Wherein, the standard score satisfies the relationship:

[0102]

[0103] Wherein, x is the actual value of the target structured data, μ and σ are respectively the mean value and variance of the target structured data in the customer data of all customers, Z(x) is the standard score of x; and the preset score is 3.

[0104] In an optional embodiment, the step of numerically encoding the standard numerical values ​​of the categorical data to obtain the characteristic value of each categorical data includes:

[0105] Obtain a preset value range of target category data, and assign different integer data as an index ID to each value in the preset value range, wherein the target category data is any one of all category data;

[0106] Determine whether the standard value of the target category data is within the preset value range;

[0107] If yes, the index ID corresponding to the standard value of the target category data is used as the characteristic value of the target category data; if no, the maximum value of all index IDs plus 1 is used as the characteristic value of the target category data;

[0108] Traverse all categorical data to obtain the feature values ​​of each categorical data.

[0109] For example, assuming that the target categorical data is the customer's city, and the preset value range of the customer's city includes 4 values: Shanghai, Beijing, Shenzhen and Zhengzhou, and the index ID of each value is 1, 2, 3, and 4 respectively; if the standard value of the target categorical data is Zhengzhou, the characteristic value of the target categorical data is 4, and if the standard value of the target categorical data is special characters, the characteristic value of the target categorical data is 5. The special characters are any characters outside the preset value range of the target structured data.

[0110] In an optional embodiment, the binning of the standard values ​​of the numerical data to obtain the characteristic value of each type of numerical data includes:

[0111] Obtaining a preset value range of target numerical data, wherein the target numerical data is any one of all numerical data;

[0112] Divide the preset value range into a preset number of sub-ranges according to a binning function, and assign different integer data as an index ID to each sub-range;

[0113] Acquire a sub-range corresponding to the standard value of the target numerical data as a target sub-range, and use the index ID of the target sub-range as a characteristic value of the target numerical data;

[0114] Traverse all numerical data to obtain the characteristic value of each numerical data.

[0115] The binning function is a cut function or a qcut function.

[0116] Exemplarily, assuming that the target numerical data is the browsing time on a preset system path, since the preset time period is one week, the preset value range of the browsing time is a continuous value from 0h (hour) to 168h (hour), and the preset value range is divided into 5 sub-ranges: [0h, 7h], (7h, 21h], (21h, 42h], (42h, 84h], (84h, 168h], and the index ID of each sub-range is 1, 2, 3, 4, 5 respectively; if the standard value of the target numerical data is 20 hours, the characteristic value of the target numerical data is 2.

[0117] In an optional embodiment, the characteristic values ​​of all categorical data and the characteristic values ​​of all numerical data are arranged together in a fixed order to construct the first customer characteristic of the target customer.

[0118] In this way, features are extracted from all structured data to obtain the first customer features of the target customers, providing a data basis for accurate recommendations.

[0119] S12: construct a second customer feature of the target customer based on the unstructured data.

[0120] In an optional embodiment, constructing the second customer feature of the target customer based on the unstructured data includes:

[0121] Performing unstructured preprocessing on the unstructured data to obtain standard sequence data, wherein the unstructured preprocessing includes serialization processing and removing preset characters;

[0122] Splitting the standard sequence data according to a preset rule to obtain a plurality of standard subsequences;

[0123] All standard subsequences are sequentially input into the trained intention recognition model to obtain the intention recognition result of each standard subsequence, and the intention recognition results of all standard subsequences are arranged according to the order of the standard subsequences in the standard sequence data to obtain an intention sequence, wherein the intention recognition result is the customer's demand degree level for the target behavior;

[0124] A second customer profile of the target customer is constructed based on the intention sequence.

[0125] In this optional embodiment, the unstructured data is the communication text between the target customer and the staff within a preset time period; the serialization process is to arrange the communication text in the order of timestamps along the row direction to obtain sequence data; the removal of preset characters is to remove stop words, useless punctuation marks, emoticons and other preset characters in the sequence data to obtain standard sequence data. The useless punctuation marks include all punctuation marks except period, exclamation mark and question mark.

[0126] In this optional embodiment, the period, exclamation mark and question mark in the standard sequence data are used as splitting points, and the sequence data between adjacent splitting points are used as a standard subsequence, and the standard subsequence is a sentence in the communication text.

[0127] In this optional embodiment, the input of the trained intent recognition model is a standard subsequence, and the output is the intent recognition result of the standard subsequence. The number of different intent recognition results is related to the business needs of the marketing scenario. After obtaining the intent recognition results of all standard subsequences, the intent recognition results of all standard subsequences are arranged according to the order of the standard subsequences in the standard sequence data to obtain an intent sequence.

[0128] In this optional embodiment, the training process of the intent recognition model includes:

[0129] Build an intent recognition model, which can be any one of LSTM, GRU, RNN, and Bert;

[0130] Collect all standard subsequences in all customers' historical communication texts, and annotate the intention recognition result for each standard subsequence to obtain multiple sets of training sequences, wherein the training sequences include the standard subsequences and the intention recognition results of the standard subsequences;

[0131] The intent recognition model is trained based on all training sequences and a cross entropy loss function to obtain a trained intent recognition model.

[0132] In an optional embodiment, constructing the second customer feature of the target customer based on the intention sequence includes:

[0133] Acquire the number of types of different intent recognition results in the intent sequence as a first statistical feature;

[0134] Obtaining the intention recognition result that appears most frequently in the intention sequence as a second statistical feature;

[0135] Taking the last intention recognition result in the intention sequence as a time feature;

[0136] A preset number of intention recognition results at the beginning of the intention sequence are intercepted as a starting sequence, and a preset number of intention recognition results at the end of the intention sequence are intercepted as an end sequence;

[0137] Calculating the difference between the proportion of each intention recognition result in the starting sequence and the ending sequence to obtain the trend characteristics of each intention recognition result;

[0138] The first statistical feature, the second statistical feature, the time feature and the trend feature of each intention recognition result are arranged in a fixed order to obtain the second customer feature of the target customer.

[0139] The preset number is 4, that is, the starting sequence and the ending sequence both include 4 intention recognition results.

[0140] Exemplarily, in the precise recommendation scenario, assuming that according to the business needs of the marketing scenario, 5 different intent recognition results A, B, C, D, and E are set, and assuming that the intent sequence is [A, B, C, A, A, C, D], the first statistical feature is 4, the second statistical feature is A, and the time feature is D; since the preset number is 4, the starting sequence and the ending sequence are [A, B, C, A] and [A, A, C, D] respectively; the trend characteristics of different intent recognition results can be obtained by calculating the difference between the proportion of each intent recognition result in the starting sequence and the ending sequence, and the trend characteristics of the intent recognition results A, B, C, D, and E are 0, 1, 0, -1, 0 respectively; the second customer characteristic of the target customer is [4, A, D, 0, 1, 0, -1, 0].

[0141] In this way, feature extraction is performed on unstructured data to obtain the second customer feature of the target customer. The second customer feature can effectively characterize the customer's demand for the product and provide a data basis for achieving accurate recommendations.

[0142] S13, inputting the first customer feature and the second customer feature into the trained precise matching model to obtain the target behavior matching rate of the target customer.

[0143] In some optional embodiments, the target behavior matching rate of the target customers can be obtained for different target behaviors according to different application scenarios. For example, in a precision marketing scenario, the target behavior corresponds to purchasing a preset product; in a precision service scenario, the target behavior corresponds to serving a preset service object; in a precision allocation of work tasks scenario, the target behavior corresponds to completing a preset work.

[0144] In an optional embodiment, inputting the first customer feature and the second customer feature into a trained accurate matching model to obtain the target behavior matching rate of the target customer includes:

[0145] Filling zeros at the ends of the first customer feature and the second customer feature to obtain the first customer feature and the second customer feature of a preset size;

[0146] The first customer feature and the second customer feature are fused according to the fusion coefficient to obtain a fusion vector, and the fusion vector satisfies the relationship:

[0147] H=ch1+(1-c)h2

[0148] Wherein, h1 and h2 are the first customer feature and the first customer feature of the preset size respectively, c is the fusion coefficient, the value range of the fusion coefficient is [0,1], and H is the fusion vector;

[0149] The fusion vector is input into the trained precise matching model to obtain the target behavior matching rate of the target customer.

[0150] The target behavior matching rate of the target customer ranges from [0, 1], and the initial value of the fusion coefficient is 0.1.

[0151] In an optional embodiment, the training process of the precise matching model includes:

[0152] Building a precise matching model, the input of which is the customer's fusion vector, and the output of which is the customer's target behavior matching rate;

[0153] Collect the first customer features and the second customer features of all customers in the historical data, and obtain a fusion vector for each customer based on the fusion coefficient;

[0154] Collect the target behavior results of each customer in the historical data, the target behavior results include success and failure. If the target behavior result of the customer is success, the label data of the customer is recorded as 1; if the target behavior result of the customer is failure, the label data of the customer is recorded as 0;

[0155] The fused vector and label data of the same customer are taken as a group of training samples, and all the training samples are stored to obtain a training data set;

[0156] The precise matching model is trained based on the training data set and the cross entropy loss function to update the parameters of the precise matching model, and the training is stopped when the cross entropy loss function no longer changes, thereby obtaining a trained precise matching model.

[0157] Among them, the precise matching model is a binary classification model, and the precise matching model can be any one of the existing binary classification models such as LightGBM, XGBoost, random forest, etc., and this application does not impose any limitation.

[0158] In this way, the trained precise matching model comprehensively considers the first customer feature and the second customer feature of the target customer to obtain the target behavior matching rate of the target customer, thereby providing a precise quantitative indicator for achieving precise recommendation and ensuring the accuracy of precise recommendation.

[0159] S14, comparing the target behavior matching rate with a preset threshold to determine whether the target customer is a customer to be matched. If the target customer is a customer to be matched, the target customer is pushed to a staff member to obtain a target behavior result.

[0160] In an optional embodiment, comparing the target behavior matching rate with a preset threshold to determine whether the target customer is a customer to be matched, and if the target customer is a customer to be matched, pushing the target customer to a staff member to obtain a target behavior result, includes:

[0161] If the target behavior matching rate is less than a preset threshold, it means that the target customer is not a customer to be matched;

[0162] If the target behavior matching rate is not less than the preset threshold, it means that the target customer is a customer to be matched, and the target customer is pushed to the staff to obtain the target behavior result of the customer to be matched, and the target behavior result includes success and failure.

[0163] Among them, the preset threshold is 0.5.

[0164] In this way, customers with no needs can be automatically excluded, reducing the workload of staff and achieving accurate recommendations.

[0165] In other optional embodiments, the method further includes step S15.

[0166] S15, updating the precise matching model based on the target behavior result and the target behavior matching rate for the next precise recommendation.

[0167] In an optional embodiment, updating the precise matching model based on the target behavior result and the target behavior matching rate specifically includes:

[0168] If the target behavior result is successful, the fusion coefficient is adjusted multiple times according to a preset rule, and the first customer feature and the second customer feature are continuously input into the trained precise matching model to obtain target behavior matching rates of multiple adjusted fusion coefficients; the fusion coefficient corresponding to the maximum value of the target behavior matching rate is selected as the new fusion coefficient to complete the update of the precise matching model;

[0169] If the target behavior result is a failure, the label data of the target customer is recorded as 0, and the fusion vector and label data of the target customer are used as a set of new training samples to train the precise matching model, thereby completing the update of the precise matching model;

[0170] The updated precise matching model is used for the next precise recommendation.

[0171] In this optional embodiment, the adjusting the fusion coefficient multiple times according to a preset rule includes: increasing the value of the fusion coefficient by 0.05 to obtain an adjusted fusion coefficient.

[0172] In this way, during the use of the precise matching model, the precise matching model is continuously updated according to the actual target behavior results to ensure the accuracy of the precise matching model.

[0173] It can be seen from the above technical solutions that the present application comprehensively considers the structured data and unstructured data that affect the accuracy of precise recommendations, and designs different feature extraction methods for different types of data to obtain the first customer feature and the second customer feature. The first customer feature and the second customer feature are comprehensively used to train a precise matching model to screen all customers and obtain customers to be matched. At the same time, the precise matching model is updated according to the target behavior results of the customers to be matched, thereby improving the accuracy and timeliness of precise recommendations.

[0174] See also Figure 2 , Figure 2 It is a functional module diagram of a preferred embodiment of the precise recommendation device based on multivariate data fusion of the present application. The precise recommendation device 11 based on multivariate data fusion includes a collection unit 110, a first construction unit 111, a second construction unit 112, an acquisition unit 113, a push unit 114, and an update unit 115. The module / unit referred to in the present application refers to a series of computer-readable instruction segments that can be executed by the processor 13 and can complete fixed functions, which are stored in the memory 12. In this embodiment, the functions of each module / unit will be described in detail in subsequent embodiments.

[0175] In an optional embodiment, the collection unit 110 is used to collect customer data of target customers within a preset time period, where the customer data includes structured data and unstructured data.

[0176] In an optional embodiment, collecting customer data of target customers within a preset time period includes:

[0177] Collecting structured data of target customers within a preset time period, wherein the structured data includes basic information of the target customers and behavior data on a preset system path, and the target customer is any one of all customers;

[0178] Collecting unstructured data of the target customer within a preset time period, wherein the unstructured data includes communication texts between the staff and the target customer;

[0179] The structured data and the unstructured data are stored to obtain customer data of the target customer.

[0180] Wherein, the preset time period is one week.

[0181] In this optional embodiment, the basic information of the target customer includes but is not limited to age, gender, occupation and the city where the customer is located; the behavioral data on the preset system path includes but is not limited to browsing time and number of clicks.

[0182] For example, in a precision marketing recommendation scenario, assuming that the target behavior is a marketing behavior of a financial product in the financial marketing industry, the preset system path is a software system including financial management apps, financial management web pages, and financial management public accounts related to the financial management products. By collecting all structured data and unstructured data of the target customer within a week, the customer data of the target customer can be obtained.

[0183] In an optional embodiment, the first constructing unit 111 is used to construct a first customer feature of the target customer based on the structured data.

[0184] In an optional embodiment, constructing the first customer feature of the target customer based on the structured data includes:

[0185] Classifying each type of structured data based on a preset value range of the structured data to obtain a classification result, wherein the classification result includes numerical data and categorical data;

[0186] Based on the classification results, all structured data are subjected to structured preprocessing to obtain a standard value of each type of structured data;

[0187] Numerical encoding is performed on the standard numerical values ​​of the categorical data to obtain characteristic values ​​of each categorical data;

[0188] Binning the standard values ​​of the numerical data to obtain characteristic values ​​of each type of numerical data;

[0189] All feature values ​​are arranged in a fixed order to construct the first customer feature of the target customer.

[0190] In an optional embodiment, classifying each type of structured data based on a preset value range of the structured data to obtain a classification result includes:

[0191] If the preset value range of the structured data is continuous, the structured data is numerical data;

[0192] If the preset value range of the structured data is discrete, the structured data is categorical data.

[0193] The preset value range corresponds to the structured data one by one.

[0194] For example, the preset value range of the city where the customer is located in the basic information of the target customer includes the names of all cities, so the city where the customer is located is categorical data; the preset value range of the browsing time in the behavioral data on the preset system path is continuous, so the browsing time is numerical data.

[0195] In an optional embodiment, the specific methods of structured preprocessing of numerical data and categorical data are different, and the structured preprocessing of all structured data based on the classification result to obtain a standard value of each type of structured data includes:

[0196] Randomly select a type of structured data as the target structured data;

[0197] If the classification result of the target structured data is numerical data, a standard score is calculated according to the actual value of the target structured data; if the standard score is not greater than the preset score, the actual value is used as the standard numerical value of the target structured data; if the standard score is greater than the preset score, the average value of the target structured data in the customer data of all customers is used as the standard numerical value of the target structured data;

[0198] If the classification result of the target structured data is categorical data, determine whether the actual value of the target structured data is within the corresponding preset value range; if so, use the actual value as the standard value of the target structured data; if not, generate a special character as the standard value of the target structured data, and the special character is any character outside the preset value range of the target structured data;

[0199] Traverse all structured data to obtain the standard value of each structured data.

[0200] Wherein, the standard score satisfies the relationship:

[0201]

[0202] Wherein, x is the actual value of the target structured data, μ and σ are respectively the mean value and variance of the target structured data in the customer data of all customers, Z(x) is the standard score of x; and the preset score is 3.

[0203] In an optional embodiment, the step of numerically encoding the standard numerical values ​​of the categorical data to obtain the characteristic value of each categorical data includes:

[0204] Obtain a preset value range of target category data, and assign different integer data as an index ID to each value in the preset value range, wherein the target category data is any one of all category data;

[0205] Determine whether the standard value of the target category data is within the preset value range;

[0206] If yes, the index ID corresponding to the standard value of the target category data is used as the characteristic value of the target category data; if no, the maximum value of all index IDs plus 1 is used as the characteristic value of the target category data;

[0207] Traverse all categorical data to obtain the feature values ​​of each categorical data.

[0208] For example, assuming that the target categorical data is the customer's city, and the preset value range of the customer's city includes 4 values: Shanghai, Beijing, Shenzhen and Zhengzhou, and the index ID of each value is 1, 2, 3, and 4 respectively; if the standard value of the target categorical data is Zhengzhou, the characteristic value of the target categorical data is 4, and if the standard value of the target categorical data is special characters, the characteristic value of the target categorical data is 5. The special characters are any characters outside the preset value range of the target structured data.

[0209] In an optional embodiment, the binning of the standard values ​​of the numerical data to obtain the characteristic value of each type of numerical data includes:

[0210] Obtaining a preset value range of target numerical data, wherein the target numerical data is any one of all numerical data;

[0211] Divide the preset value range into a preset number of sub-ranges according to a binning function, and assign different integer data as an index ID to each sub-range;

[0212] Acquire a sub-range corresponding to the standard value of the target numerical data as a target sub-range, and use the index ID of the target sub-range as a characteristic value of the target numerical data;

[0213] Traverse all numerical data to obtain the characteristic value of each numerical data.

[0214] The binning function is a cut function or a qcut function.

[0215] Exemplarily, assuming that the target numerical data is the browsing time on a preset system path, since the preset time period is one week, the preset value range of the browsing time is a continuous value from 0h (hour) to 168h (hour), and the preset value range is divided into 5 sub-ranges: [0h, 7h], (7h, 21h], (21h, 42h], (42h, 84h], (84h, 168h], and the index ID of each sub-range is 1, 2, 3, 4, 5 respectively; if the standard value of the target numerical data is 20 hours, the characteristic value of the target numerical data is 2.

[0216] In an optional embodiment, the characteristic values ​​of all categorical data and the characteristic values ​​of all numerical data are arranged together in a fixed order to construct the first customer characteristic of the target customer.

[0217] In an optional embodiment, the second constructing unit 112 is configured to construct a second customer feature of the target customer based on the unstructured data.

[0218] In an optional embodiment, constructing the second customer feature of the target customer based on the unstructured data includes:

[0219] Performing unstructured preprocessing on the unstructured data to obtain standard sequence data, wherein the unstructured preprocessing includes serialization processing and removing preset characters;

[0220] Splitting the standard sequence data according to a preset rule to obtain a plurality of standard subsequences;

[0221] All standard subsequences are sequentially input into the trained intention recognition model to obtain the intention recognition result of each standard subsequence, and the intention recognition results of all standard subsequences are arranged according to the order of the standard subsequences in the standard sequence data to obtain an intention sequence, wherein the intention recognition result is the customer's demand degree level for the target behavior;

[0222] A second customer profile of the target customer is constructed based on the intention sequence.

[0223] In this optional embodiment, the unstructured data is the communication text between the target customer and the staff within a preset time period; the serialization process is to arrange the communication text in the order of timestamps along the row direction to obtain sequence data; the removal of preset characters is to remove stop words, useless punctuation marks, emoticons and other preset characters in the sequence data to obtain standard sequence data. The useless punctuation marks include all punctuation marks except period, exclamation mark and question mark.

[0224] In this optional embodiment, the period, exclamation mark and question mark in the standard sequence data are used as splitting points, and the sequence data between adjacent splitting points are used as a standard subsequence, and the standard subsequence is a sentence in the communication text.

[0225] In this optional embodiment, the input of the trained intent recognition model is a standard subsequence, and the output is the intent recognition result of the standard subsequence. The number of different intent recognition results is related to the business needs of the marketing scenario. After obtaining the intent recognition results of all standard subsequences, the intent recognition results of all standard subsequences are arranged according to the order of the standard subsequences in the standard sequence data to obtain an intent sequence.

[0226] In this optional embodiment, the training process of the intent recognition model includes:

[0227] Build an intent recognition model, which can be any one of LSTM, GRU, RNN, and Bert;

[0228] Collect all standard subsequences in all customers' historical communication texts, and annotate the intention recognition result for each standard subsequence to obtain multiple sets of training sequences, wherein the training sequences include the standard subsequences and the intention recognition results of the standard subsequences;

[0229] The intent recognition model is trained based on all training sequences and a cross entropy loss function to obtain a trained intent recognition model.

[0230] In an optional embodiment, constructing the second customer feature of the target customer based on the intention sequence includes:

[0231] Acquire the number of types of different intent recognition results in the intent sequence as a first statistical feature;

[0232] Obtaining the intention recognition result that appears most frequently in the intention sequence as a second statistical feature;

[0233] Taking the last intention recognition result in the intention sequence as a time feature;

[0234] A preset number of intention recognition results at the beginning of the intention sequence are intercepted as a starting sequence, and a preset number of intention recognition results at the end of the intention sequence are intercepted as an end sequence;

[0235] Calculating the difference between the proportion of each intention recognition result in the starting sequence and the ending sequence to obtain the trend characteristics of each intention recognition result;

[0236] The first statistical feature, the second statistical feature, the time feature and the trend feature of each intention recognition result are arranged in a fixed order to obtain the second customer feature of the target customer.

[0237] The preset number is 4, that is, the starting sequence and the ending sequence both include 4 intention recognition results.

[0238] Exemplarily, in the precise recommendation scenario, assuming that according to the business needs of the marketing scenario, 5 different intent recognition results A, B, C, D, and E are set, and assuming that the intent sequence is [A, B, C, A, A, C, D], the first statistical feature is 4, the second statistical feature is A, and the time feature is D; since the preset number is 4, the starting sequence and the ending sequence are [A, B, C, A] and [A, A, C, D] respectively; the trend characteristics of different intent recognition results can be obtained by calculating the difference between the proportion of each intent recognition result in the starting sequence and the ending sequence, and the trend characteristics of the intent recognition results A, B, C, D, and E are 0, 1, 0, -1, 0 respectively; the second customer characteristic of the target customer is [4, A, D, 0, 1, 0, -1, 0].

[0239] In an optional embodiment, the acquisition unit 113 is used to input the first customer feature and the second customer feature into a trained accurate matching model to obtain a target behavior matching rate of the target customer.

[0240] In some optional embodiments, the target behavior matching rate of the target customers can be obtained for different target behaviors according to different application scenarios. For example, in a precision marketing scenario, the target behavior corresponds to purchasing a preset product; in a precision service scenario, the target behavior corresponds to serving a preset service object; in a precision allocation of work tasks scenario, the target behavior corresponds to completing a preset work.

[0241] In an optional embodiment, inputting the first customer feature and the second customer feature into a trained accurate matching model to obtain the target behavior matching rate of the target customer includes:

[0242] Filling zeros at the ends of the first customer feature and the second customer feature to obtain the first customer feature and the second customer feature of a preset size;

[0243] The first customer feature and the second customer feature are fused according to the fusion coefficient to obtain a fusion vector, and the fusion vector satisfies the relationship:

[0244] H=ch1+(1-c)h2

[0245] Wherein, h1 and h2 are the first customer feature and the first customer feature of the preset size respectively, c is the fusion coefficient, the value range of the fusion coefficient is [0,1], and H is the fusion vector;

[0246] The fusion vector is input into the trained precise matching model to obtain the target behavior matching rate of the target customer.

[0247] The target behavior matching rate of the target customer ranges from [0, 1], and the initial value of the fusion coefficient is 0.1.

[0248] In an optional embodiment, the training process of the precise matching model includes:

[0249] Building a precise matching model, the input of which is the customer's fusion vector, and the output of which is the customer's target behavior matching rate;

[0250] Collect the first customer features and the second customer features of all customers in the historical data, and obtain a fusion vector for each customer based on the fusion coefficient;

[0251] Collect the target behavior results of each customer in the historical data, the target behavior results include success and failure. If the target behavior result of the customer is success, the label data of the customer is recorded as 1; if the target behavior result of the customer is failure, the label data of the customer is recorded as 0;

[0252] The fused vector and label data of the same customer are taken as a group of training samples, and all the training samples are stored to obtain a training data set;

[0253] The precise matching model is trained based on the training data set and the cross entropy loss function to update the parameters of the precise matching model, and the training is stopped when the cross entropy loss function no longer changes, thereby obtaining a trained precise matching model.

[0254] Among them, the precise matching model is a binary classification model, and the precise matching model can be any one of the existing binary classification models such as LightGBM, XGBoost, random forest, etc., and this application does not impose any limitation.

[0255] In an optional embodiment, the push unit 114 is used to compare the target behavior matching rate with a preset threshold to determine whether the target customer is a customer to be matched. If the target customer is a customer to be matched, the target customer is pushed to the staff to obtain the target behavior result.

[0256] In an optional embodiment, comparing the target behavior matching rate with a preset threshold to determine whether the target customer is a customer to be matched, and if the target customer is a customer to be matched, pushing the target customer to a staff member to obtain a target behavior result, includes:

[0257] If the target behavior matching rate is less than a preset threshold, it means that the target customer is not a customer to be matched;

[0258] If the target behavior matching rate is not less than the preset threshold, it means that the target customer is a customer to be matched, and the target customer is pushed to the staff to obtain the target behavior result of the customer to be matched, and the target behavior result includes success and failure. Among them, the preset threshold is 0.5.

[0259] In other optional embodiments, the device further includes an updating unit 115. The updating unit 115 is used to update the precise matching model based on the target behavior result and the target behavior matching rate for the next precise recommendation.

[0260] In an optional embodiment, updating the precise matching model based on the target behavior result and the target behavior matching rate specifically includes:

[0261] If the target behavior result is successful, the fusion coefficient is adjusted multiple times according to a preset rule, and the first customer feature and the second customer feature are continuously input into the trained precise matching model to obtain target behavior matching rates of multiple adjusted fusion coefficients; the fusion coefficient corresponding to the maximum value of the target behavior matching rate is selected as the new fusion coefficient to complete the update of the precise matching model;

[0262] If the target behavior result is a failure, the label data of the target customer is recorded as 0, and the fusion vector and label data of the target customer are used as a set of new training samples to train the precise matching model, thereby completing the update of the precise matching model;

[0263] The updated precise matching model is used for the next precise recommendation.

[0264] In this optional embodiment, the adjusting the fusion coefficient multiple times according to a preset rule includes: increasing the value of the fusion coefficient by 0.05 to obtain an adjusted fusion coefficient.

[0265] It can be seen from the above technical solutions that the present application comprehensively considers the structured data and unstructured data that affect the accuracy of precise recommendations, and designs different feature extraction methods for different types of data to obtain the first customer feature and the second customer feature. The first customer feature and the second customer feature are comprehensively used to train a precise matching model to screen all customers and obtain customers to be matched. At the same time, the precise matching model is updated according to the target behavior results of the customers to be matched, thereby improving the accuracy and timeliness of precise recommendations.

[0266] See also Figure 3 , is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device 1 includes a memory 12 and a processor 13. The memory 12 is used to store computer-readable instructions, and the processor 13 is used to execute the computer-readable instructions stored in the memory to implement the accurate recommendation method based on multivariate data fusion described in any of the above embodiments.

[0267] In an optional embodiment, the electronic device 1 further includes a bus, and a computer program stored in the memory 12 and executable on the processor 13, such as an accurate recommendation program based on multivariate data fusion.

[0268] Figure 3 Only the electronic device 1 having the memory 12 and the processor 13 is shown, and those skilled in the art can understand that Figure 3 The structure shown does not constitute a limitation on the electronic device 1 , and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0269] Combination Figure 1 The memory 12 in the electronic device 1 stores a plurality of computer-readable instructions to implement an accurate recommendation method based on multivariate data fusion, and the processor 13 can execute the plurality of instructions to implement:

[0270] Collecting customer data of target customers within a preset time period, wherein the customer data includes structured data and unstructured data;

[0271] constructing a first customer feature of the target customer based on the structured data;

[0272] constructing a second customer feature of the target customer based on the unstructured data;

[0273] Inputting the first customer feature and the second customer feature into the trained precise matching model to obtain the target behavior matching rate of the target customer;

[0274] The target behavior matching rate is compared with a preset threshold to determine whether the target customer is a customer to be matched. If the target customer is a customer to be matched, the target customer is pushed to the staff to obtain the target behavior result.

[0275] Specifically, the specific implementation method of the processor 13 for the above instructions can refer to Figure 1 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0276] Those skilled in the art will appreciate that the schematic diagram is merely an example of the electronic device 1 and does not constitute a limitation on the electronic device 1. The electronic device 1 may be a bus structure or a star structure. The electronic device 1 may also include more or less other hardware or software than shown in the diagram, or a different arrangement of components. For example, the electronic device 1 may also include input and output devices, network access devices, etc.

[0277] It should be noted that the electronic device 1 is only an example, and other existing or future electronic products that are suitable for the present application should also be included in the protection scope of the present application and included here by reference.

[0278] Among them, the memory 12 includes at least one type of readable storage medium, and the readable storage medium can be non-volatile or volatile. The readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 12 can be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 12 can also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. The memory 12 can not only be used to store application software and various types of data installed on the electronic device 1, such as the code of the accurate recommendation program based on multivariate data fusion, but also can be used to temporarily store data that has been output or is to be output.

[0279] In some embodiments, the processor 13 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and a combination of various control chips. The processor 13 is the control core (Control Unit) of the electronic device 1, and uses various interfaces and lines to connect the various components of the entire electronic device 1, and executes or executes programs or modules stored in the memory 12 (for example, executing a precise recommendation program based on multivariate data fusion, etc.), and calls the data stored in the memory 12 to execute various functions of the electronic device 1 and process data.

[0280] The processor 13 executes the operating system of the electronic device 1 and various installed applications. The processor 13 executes the applications to implement the steps in the above-mentioned embodiments of the precise recommendation method based on multivariate data fusion, for example Figure 1 Steps shown.

[0281] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory 12 and executed by the processor 13 to complete the present application. The one or more modules / units may be a series of computer-readable instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program in the electronic device 1. For example, the computer program may be divided into a collection unit 110, a first construction unit 111, a second construction unit 112, an acquisition unit 113, a push unit 114, and an update unit 115.

[0282] The above-mentioned integrated unit implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a computer device, or a network device, etc.) or a processor to execute the part of the precise recommendation method based on multivariate data fusion described in each embodiment of the present application.

[0283] If the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also instruct the relevant hardware devices to complete through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, the steps of each of the above-mentioned method embodiments can be implemented.

[0284] The computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory and other memory, etc.

[0285] Furthermore, the computer-readable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the blockchain node, etc.

[0286] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0287] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. Figure 3 Only one arrow is used in the figure, but it does not mean that there is only one bus or one type of bus. The bus is configured to realize the connection and communication between the memory 12 and at least one processor 13, etc.

[0288] An embodiment of the present application also provides a computer-readable storage medium (not shown), in which computer-readable storage medium is stored computer-readable instructions, and the computer-readable instructions are executed by a processor in an electronic device to implement the precise recommendation method based on multivariate data fusion described in any of the above embodiments.

[0289] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0290] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0291] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0292] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the specification can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any specific order.

[0293] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present application and are not intended to limit it. Although the present application has been described in detail with reference to the preferred embodiments, a person of ordinary skill in the art should understand that the technical solution of the present application may be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present application.

Claims

1. An accurate recommendation method based on multivariate data fusion, characterized in that: The method comprises: Collecting customer data of target customers within a preset time period, wherein the customer data includes structured data and unstructured data; Constructing the first customer feature of the target customer based on the structured data, including: classifying the structured data into numerical data and categorical data according to a preset value range; performing numerical encoding conversion on the standardized categorical data to obtain corresponding feature values, performing binning processing on the standardized numerical data to obtain corresponding feature values; arranging all feature values ​​in a fixed dimensional order to construct the first customer feature; Constructing the second customer feature of the target customer based on the unstructured data, including: performing unstructured preprocessing on the unstructured data to obtain standard sequence data; splitting the standard sequence data according to preset rules to obtain multiple standard subsequences; inputting all the standard subsequences into the trained intention recognition model in sequence to obtain the intention recognition result of each standard subsequence, and arranging the intention recognition results of all the standard subsequences according to the order of the standard subsequences in the standard sequence data to obtain an intention sequence, wherein the intention recognition result is the customer's demand degree level for the target behavior; constructing the second customer feature of the target customer based on the intention sequence; Inputting the first customer feature and the second customer feature into the trained precise matching model to obtain the target behavior matching rate of the target customer, including: filling 0 at the ends of the first customer feature and the second customer feature to obtain the first customer feature and the second customer feature of a preset size; fusing the first customer feature and the second customer feature according to a fusion coefficient to obtain a fusion vector; and inputting the fusion vector into the trained precise matching model to obtain the target behavior matching rate of the target customer; The target behavior matching rate is compared with a preset threshold to determine whether the target customer is a customer to be matched. If the target customer is a customer to be matched, the target customer is pushed to the staff to obtain the target behavior result.

2. The accurate recommendation method based on multivariate data fusion according to claim 1, characterized in that: The collecting of customer data of target customers within a preset time period includes: Collecting structured data of target customers within a preset time period, wherein the structured data at least includes basic information of the target customers and behavior data on a preset system path, and the target customer is any one of all customers; Collecting unstructured data of the target customer within a preset time period, wherein the unstructured data at least includes communication texts between the staff and the target customer; The structured data and the unstructured data are stored to obtain customer data of the target customer.

3. The accurate recommendation method based on multivariate data fusion according to claim 1, characterized in that: The step of constructing the first customer feature of the target customer based on the structured data includes: Classifying each type of structured data based on a preset value range of the structured data to obtain a classification result, wherein the classification result includes numerical data and categorical data; Based on the classification results, all structured data are subjected to structured preprocessing to obtain a standard value of each type of structured data; Numerical encoding is performed on the standard numerical values ​​of the categorical data to obtain characteristic values ​​of each categorical data; Binning the standard values ​​of the numerical data to obtain characteristic values ​​of each type of numerical data; All feature values ​​are arranged in a fixed order to construct the first customer feature of the target customer.

4. The accurate recommendation method based on multivariate data fusion according to claim 3, characterized in that: The structured preprocessing of all structured data based on the classification result to obtain a standard value of each type of structured data includes: Randomly select a type of structured data as the target structured data; If the classification result of the target structured data is numerical data, a standard score is calculated according to the actual value of the target structured data; if the standard score is not greater than the preset score, the actual value is used as the standard numerical value of the target structured data; if the standard score is greater than the preset score, the average value of the target structured data in the customer data of all customers is used as the standard numerical value of the target structured data; If the classification result of the target structured data is categorical data, determine whether the actual value of the target structured data is within the corresponding preset value range; if so, use the actual value as the standard value of the target structured data; if not, generate a special character as the standard value of the target structured data, and the special character is any character outside the preset value range of the target structured data; Traverse all structured data to obtain the standard value of each structured data.

5. The accurate recommendation method based on multivariate data fusion according to claim 3, characterized in that: The binning process of the standard values ​​of the numerical data to obtain the characteristic values ​​of each type of numerical data includes: Obtaining a preset value range of target numerical data, wherein the target numerical data is any one of all numerical data; Divide the preset value range into a preset number of sub-ranges according to a binning function, and assign different integer data as an index ID to each sub-range; Acquire a sub-range corresponding to the standard value of the target numerical data as a target sub-range, and use the index ID of the target sub-range as a characteristic value of the target numerical data; Traverse all numerical data to obtain the characteristic value of each numerical data.

6. The accurate recommendation method based on multivariate data fusion according to claim 1, characterized in that: The unstructured preprocessing includes serialization processing and removing preset characters.

7. The accurate recommendation method based on multivariate data fusion according to claim 6, characterized in that: The constructing the second customer feature of the target customer based on the intention sequence includes: Acquire the number of types of different intent recognition results in the intent sequence as a first statistical feature; Obtaining the intention recognition result that appears most frequently in the intention sequence as a second statistical feature; Taking the last intention recognition result in the intention sequence as a time feature; A preset number of intention recognition results at the beginning of the intention sequence are intercepted as a starting sequence, and a preset number of intention recognition results at the end of the intention sequence are intercepted as an end sequence; Calculating the difference between the proportion of each intention recognition result in the starting sequence and the ending sequence to obtain the trend characteristics of each intention recognition result; The first statistical feature, the second statistical feature, the time feature and the trend feature of each intention recognition result are arranged in a fixed order to obtain the second customer feature of the target customer.

8. The accurate recommendation method based on multivariate data fusion according to claim 1, characterized in that: The fusion vector satisfies the relationship: in, and are the first customer feature and the first customer feature of the preset size respectively, c is the fusion coefficient, and the value range of the fusion coefficient is [0,1], is the fusion vector.

9. The accurate recommendation method based on multivariate data fusion according to claim 8, characterized in that: The method further includes: updating the precise matching model based on the target behavior result and the target behavior matching rate, specifically including: If the target behavior result is successful, the fusion coefficient is adjusted multiple times according to a preset rule, and the first customer feature and the second customer feature are continuously input into the trained precise matching model to obtain target behavior matching rates of multiple adjusted fusion coefficients; the fusion coefficient corresponding to the maximum value of the target behavior matching rate is selected as the new fusion coefficient to complete the update of the precise matching model; If the target behavior result is a failure, the label data of the target customer is recorded as 0, and the fusion vector and label data of the target customer are used as a set of new training samples to train the precise matching model, thereby completing the update of the precise matching model; The updated precise matching model is used for the next precise recommendation.

10. A precise recommendation device based on multivariate data fusion for implementing the method according to claim 1, characterized in that: The device comprises: A collection unit, used to collect customer data of target customers within a preset time period, wherein the customer data includes structured data and unstructured data; A first construction unit, configured to construct a first customer feature of the target customer based on the structured data; A second construction unit, configured to construct a second customer feature of the target customer based on the unstructured data; An acquisition unit, configured to input the first customer feature and the second customer feature into a trained accurate matching model to acquire a target behavior matching rate of the target customer; The push unit is used to compare the target behavior matching rate with a preset threshold to determine whether the target customer is a customer to be matched. If the target customer is a customer to be matched, the target customer is pushed to the staff to obtain the target behavior result.

11. An electronic device, characterized in that: The electronic device comprises: a memory storing computer-readable instructions; and A processor executes computer-readable instructions stored in the memory to implement the accurate recommendation method based on multivariate data fusion as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the accurate recommendation method based on multivariate data fusion as described in any one of claims 1 to 9 is implemented.

Images