Recommendation generation method and device based on static and dynamic data, equipment and medium

Abstract

The application relates to the field of financial technology, and discloses a recommendation generation method based on static and dynamic data, which comprises the following steps: obtaining static basic information and dynamic behavior data of a target user, converting the static basic information into a static vector matrix, classifying the dynamic behavior data, calculating the frequency value and trend change coefficient of the dynamic behavior data under each classification label, converting the dynamic behavior data into a dynamic vector matrix according to the classification label, the frequency value and the trend change coefficient, weighting the static vector matrix and the dynamic vector matrix respectively by using a preset attention mechanism, fusing the weighted static vector matrix and the weighted dynamic vector matrix, and generating a result recommended to the target user according to a fused vector matrix by using a pre-trained recommendation model. The application further discloses a recommendation generation device based on static and dynamic data, an electrical device and a medium. The application can improve the generation accuracy of intelligent recommendation in the financial field.

Description

Technical Field

[0001] This invention relates to the field of financial technology, and in particular to a method, apparatus, electronic device, and computer-readable storage medium for generating recommendations based on static and dynamic data. Background Technology

[0002] Intelligent recommendation systems are widely used in the financial sector, for example, in recommending wealth management products such as funds and bonds, insurance services, and financial news.

[0003] Traditional intelligent recommendation systems mostly employ content-based recommendation algorithms, meaning the system recommends content that a user might be interested in based on their past behavior. This approach, because it doesn't further segment and analyze past user behavior data, results in relatively homogenous and fixed content recommendations. It fails to adapt to changes in user interests in real time, leading to repeated exposure to the same types of content, which can cause user boredom and negatively impact the user experience. Summary of the Invention

[0004] This invention provides a recommendation generation method, apparatus, electronic device, and computer-readable storage medium based on static and dynamic data, with the main objective of improving the accuracy of intelligent recommendation generation in the financial field.

[0005] To achieve the above objectives, the present invention provides a recommendation generation method based on static and dynamic data, comprising:

[0006] Obtain static basic information and dynamic behavioral data of the target user;

[0007] Convert the static basic information into a static vector matrix;

[0008] The dynamic behavior data is classified to determine the corresponding classification labels, and the frequency value and trend change coefficient of the dynamic behavior data under each classification label are calculated.

[0009] Based on the classification labels, frequency values, and trend change coefficients, the corresponding dynamic behavior data is converted into a dynamic vector matrix;

[0010] The static vector matrix and the dynamic vector matrix are weighted separately using a preset attention mechanism, and the weighted static vector matrix and the weighted dynamic vector matrix are fused to obtain a fused vector matrix.

[0011] Using a pre-trained recommendation model, recommendations are generated for the target user based on the fusion vector matrix.

[0012] Optionally, converting the static basic information into a static vector matrix includes:

[0013] The static basic information is semantically divided to obtain one or more static information units;

[0014] Extract keywords from each static information unit, and perform fuzzy matching between the keywords and preset static information tags to obtain the static information tags corresponding to each static information unit;

[0015] Obtain the value range corresponding to each static information tag, and convert the corresponding static information unit into a static numerical point using the segmented conversion method according to the value range.

[0016] Perform vector transformation operations on the static numerical points corresponding to all static information units to obtain the static vector matrix.

[0017] Optionally, classifying the dynamic behavior data and determining the corresponding classification labels for the dynamic behavior data includes:

[0018] Extract the data features of each piece of dynamic behavior data and the tag features of each preset dynamic behavior tag in sequence;

[0019] The similarity between the data features of each piece of dynamic behavior data and each of the label features is calculated sequentially.

[0020] The dynamic behavior label corresponding to the maximum similarity is selected as the classification label of the corresponding dynamic behavior data.

[0021] Optionally, calculating the frequency value and trend change coefficient of dynamic behavior data under each category label includes:

[0022] Using dynamic behavior data under the same category label as the target object, obtain the timestamp set of the target object;

[0023] From the set of timestamps, remove the smallest timestamp and randomly select two timestamps a preset number of times as the measurement point timestamps;

[0024] Calculate the difference in frequency values ​​of dynamic behavior data under the same category label for each of the two timestamps of the measurement points relative to the minimum timestamp;

[0025] When the difference between the frequency values ​​is greater than 0, the count is 1; when the difference between the frequency values ​​is less than 0, the count is -1; when the difference between the frequency values ​​is equal to 0, the count is 0.

[0026] Calculate the sum of counts corresponding to each category label within the preset number of times. If the sum of counts is greater than 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to 1. If the sum of counts is equal to 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to 0. If the sum of counts is less than 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to -1.

[0027] Optionally, the step of converting the corresponding dynamic behavior data into a dynamic vector matrix based on the classification label, frequency value, and trend change coefficient includes:

[0028] The category label, frequency value and trend change coefficient of the dynamic behavior data under the same category label are successively vectorized and concatenated to obtain the one-dimensional vector corresponding to the dynamic behavior data.

[0029] Obtain the dimension of the static vector matrix, and combine the one-dimensional vectors corresponding to all dynamic behavior data according to the dimension of the static vector matrix to obtain the dynamic vector matrix.

[0030] To address the above problems, the present invention also provides a recommendation generation device based on static and dynamic data, the device comprising:

[0031] The original data acquisition module is used to acquire the static basic information and dynamic behavior data of the target user;

[0032] A static vector conversion module is used to convert the static basic information into a static vector matrix;

[0033] The dynamic vector conversion module is used to classify the dynamic behavior data, determine the classification label corresponding to the dynamic behavior data, calculate the frequency value and trend change coefficient of the dynamic behavior data under each classification label, and convert the corresponding dynamic behavior data into a dynamic vector matrix based on the classification label, frequency value and trend change coefficient.

[0034] The static and dynamic fusion module is used to weight the static vector matrix and the dynamic vector matrix respectively using a preset attention mechanism, and then fuse the weighted static vector matrix and the weighted dynamic vector matrix to obtain a fused vector matrix.

[0035] The recommendation prediction module is used to generate recommendations to the target user based on the fusion vector matrix using a pre-trained recommendation model.

[0036] Optionally, the original data acquisition module converts the static basic information into a static vector matrix using the following method:

[0037] The static basic information is semantically divided to obtain one or more static information units;

[0038] Extract keywords from each static information unit, and perform fuzzy matching between the keywords and preset static information tags to obtain the static information tags corresponding to each static information unit;

[0039] Obtain the value range corresponding to each static information tag, and convert the corresponding static information unit into a static numerical point using the segmented conversion method according to the value range.

[0040] Perform vector transformation operations on the static numerical points corresponding to all static information units to obtain the static vector matrix.

[0041] Optionally, the dynamic vector conversion module classifies the dynamic behavior data using the following method to determine the corresponding classification label for the dynamic behavior data:

[0042] Extract the data features of each piece of dynamic behavior data and the tag features of each preset dynamic behavior tag in sequence;

[0043] The similarity between the data features of each piece of dynamic behavior data and each of the label features is calculated sequentially.

[0044] The dynamic behavior label corresponding to the maximum similarity is selected as the classification label of the corresponding dynamic behavior data.

[0045] To address the above problems, the present invention also provides an electronic device, the electronic device comprising:

[0046] Memory, storing at least one computer program; and

[0047] The processor executes the program stored in the memory to implement the recommendation generation method based on static and dynamic data described above.

[0048] To address the aforementioned problems, the present invention also provides a computer-readable storage medium storing at least one computer program, which is executed by a processor in an electronic device to implement the recommendation generation method based on static and dynamic data described above.

[0049] This invention application starts from two dimensions: static basic information and dynamic behavioral data of users. On the one hand, by calculating the frequency value and trend change coefficient of each type of dynamic behavioral data, it reflects both the differences between user dynamic behavioral data and the essence of dynamic behavioral data changes. On the other hand, by using a preset attention mechanism to weight the static vector matrix and the dynamic vector matrix respectively, the weight of static basic data and dynamic behavioral data is dynamically determined, ensuring that the final fused vector matrix can truly reflect the changing trend of user interests, making the recommendation results closer to the user's current actual needs and improving the accuracy of the recommendation. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating a recommendation generation method based on static and dynamic data according to an embodiment of the present invention.

[0051] Figure 2 This is a detailed implementation flowchart of one step in a recommendation generation method based on static and dynamic data according to an embodiment of the present invention.

[0052] Figure 3 A functional block diagram of a recommendation generation device based on static and dynamic data provided in an embodiment of the present invention;

[0053] Figure 4 This is a schematic diagram of the structure of an electronic device that implements the recommendation generation method based on static and dynamic data, according to an embodiment of the present invention.

[0054] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0055] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0056] This application provides a recommendation generation method based on static and dynamic data. The execution entity of this recommendation generation method includes, but is not limited to, at least one of the following: a server, a terminal, or other electronic devices configured to execute the method provided in this application. In other words, the recommendation generation method based on static and dynamic data can be executed by software or hardware installed on a terminal device or a server device; the software can be a blockchain platform. The server can be an independent server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0057] Reference Figure 1 The diagram shown is a flowchart illustrating a recommendation generation method based on static and dynamic data according to an embodiment of the present invention. In this embodiment, the recommendation generation method based on static and dynamic data includes:

[0058] S1. Obtain the static basic information and dynamic behavior data of the target user;

[0059] In this embodiment of the invention, the static basic information refers to the background information of the target user. This information has a certain degree of stability and is not easily changed in a short period of time. For example, the user's gender, education level, occupation, income level, city of residence, physical health status, and favorite sports.

[0060] In this embodiment of the invention, the dynamic behavioral data refers to a series of behavioral data of the target user within a specified observation period, surrounding the actual recommendation business scenario. This dynamic behavioral data has a certain degree of randomness and is prone to jumps. For example, in a recommendation scenario for insurance services, information such as the user's insurance business transactions or browsing behavior over the past two years can be collected, including purchasing a certain type of insurance, canceling a certain type of insurance, online consultation on a certain type of insurance, browsing insurance products, and complaints / reviews. Information such as medical treatment and property purchases by the user over the past two years can also be collected. It should be noted that all the information collected above is information that can be collected only after authorization from the target user.

[0061] It is understandable that the financial information that users pay attention to is closely related to their static basic information. For example, the higher a user's education level, the wider the range of financial products they may be interested in; similarly, residents of developed cities are more likely to engage in financial management than residents of ordinary cities. On the other hand, users' past dynamic behavior data can also reflect their interests. For instance, if a user frequently browses or inquires about a certain type of financial product within a short period of time, it indicates that the user is currently paying attention to that financial product. Therefore, this embodiment of the invention analyzes the content that a user may be interested in from both the user's static basic information and dynamic behavior data, ensuring the accuracy of the recommendation analysis source.

[0062] S2. Convert the static basic information into a static vector matrix;

[0063] In this embodiment of the invention, by converting the static basic information into a static vector matrix, the static basic information, which is text data composed of natural language, can be converted into vector data that can be recognized by the subsequent recommendation model based on neural network deep learning.

[0064] For details, please refer to Figure 2 As shown, S2 includes:

[0065] S21. Semantically divide the static basic information to obtain one or more static information units;

[0066] S22. Extract keywords from each static information unit, and perform fuzzy matching between the keywords and preset static information tags to obtain static information tags corresponding to each static information unit;

[0067] S23. Obtain the value range corresponding to each static information tag, and convert the corresponding static information unit into a static numerical point using the segmented conversion method according to the value range.

[0068] S24. Perform vector transformation operation on the static numerical points corresponding to all static information units to obtain the static vector matrix.

[0069] In this embodiment of the invention, an intelligent model with semantic recognition function can be used to perform semantic analysis on the static basic information, and the static basic information expressing the same semantic information can be divided into a static information unit.

[0070] In this embodiment of the invention, the preset static information tag refers to the information type used to describe the type of a certain static basic information. It can be set according to the actual business situation. For example, the preset static information tag includes, but is not limited to, tags such as gender, education level, income level, place of residence, and occupation.

[0071] In this embodiment of the invention, the value range corresponding to each static information tag can be set according to the actual situation. For example, for the static information tag of gender, the value range of the static information tag includes 1 or 0, where 1 represents male and 0 represents female.

[0072] In this embodiment of the invention, a segmented conversion method is used to convert the corresponding static information units into static numerical points. For example, if the target user has a high school education, the value range corresponding to the education level is [1-4], where 1 represents high school or below, 2 represents undergraduate, 3 represents postgraduate, and 4 represents postgraduate or above. Therefore, the static numerical point corresponding to the static basic data of the target user's education level static information tag is 1.

[0073] S3. Classify the dynamic behavior data, determine the classification labels corresponding to the dynamic behavior data, and calculate the frequency value and trend change coefficient of the dynamic behavior data under each classification label.

[0074] In this embodiment of the invention, the dynamic behavior of the target user can be classified according to actual business needs. For example, it can be distinguished by the nature of the action, including but not limited to browsing, consultation, assistance, and evaluation. It can also be classified according to the different objects involved in the same action, for example, based on the different target objects browsed, the dynamic behavior data includes but is not limited to fund browsing behavior, bond browsing behavior, and insurance browsing behavior. Furthermore, it can be distinguished according to the different means by which the action is performed, the dynamic behavior data can be classified as online consultation, telephone consultation, and on-site consultation, etc.

[0075] Understandably, the purpose of categorizing the dynamic behavioral data is to reflect the differences in user focus and level of attention among different types of dynamic behavioral data. For example, the level of user interest in an insurance product reflected by two types of behavioral data—one being a face-to-face consultation with staff at an insurance branch and the other being an online consultation with a smart customer service representative via an app—is different. Generally, face-to-face communication reflects a higher level of user interest in the product.

[0076] In this embodiment of the invention, clustering and other techniques can be used to classify the dynamic behavioral data.

[0077] In another optional embodiment of the present invention, the dynamic behavior data can be classified and the corresponding classification labels can be determined by the following method:

[0078] Extract the data features of each piece of dynamic behavior data and the tag features of each preset dynamic behavior tag in sequence;

[0079] The similarity between the data features of each piece of dynamic behavior data and each of the label features is calculated sequentially.

[0080] The dynamic behavior label corresponding to the maximum similarity is selected as the classification label of the corresponding dynamic behavior data.

[0081] In this embodiment of the invention, the preset dynamic behavior tags can be set according to actual business needs, such as online consultation tags, telephone consultation tags, and on-site consultation tags.

[0082] In this embodiment of the invention, a pre-trained machine learning model with semantic recognition function can be used to extract data features of each dynamic behavior data and label features of preset dynamic behavior tags. The data features include, but are not limited to, dynamic behavior occurrence path features, such as telephone or on-site; dynamic behavior target object features, such as insurance products, fund products, etc.; and dynamic behavior emotional features, such as evaluation terms like "very good" or "dissatisfied".

[0083] In this embodiment of the invention, the tag feature refers to the text feature of each preset dynamic behavior tag.

[0084] In this embodiment of the invention, the cosine similarity algorithm can be used to calculate the similarity between the data features and the label features.

[0085] In this embodiment of the invention, the frequency value of each type of dynamic behavior data refers to the total number of times that type of dynamic behavior occurs within the observation period, reflecting the overall activity level of that type of dynamic behavior data. The trend change coefficient of each type of dynamic behavior data refers to the activity trend exhibited by that type of dynamic behavior within the observation period, including becoming increasingly active, becoming increasingly inactive, and basically stable. For increasingly active behavior, the corresponding trend change coefficient is 1; for increasingly inactive behavior, the corresponding trend change coefficient is -1; and for basically stable behavior, the corresponding trend change coefficient is 0.

[0086] In this embodiment of the invention, the frequency value and trend change coefficient of dynamic behavior data under each category label can be calculated by the following method, including:

[0087] Using dynamic behavior data under the same category label as the target object, obtain the timestamp set of the target object;

[0088] From the set of timestamps, remove the smallest timestamp and randomly select two timestamps a preset number of times as the measurement point timestamps;

[0089] Calculate the difference in frequency values ​​of dynamic behavior data under the same category label for each of the two timestamps of the measurement points relative to the minimum timestamp;

[0090] When the difference between the frequency values ​​is greater than 0, the count is 1; when the difference between the frequency values ​​is less than 0, the count is -1; when the difference between the frequency values ​​is equal to 0, the count is 0.

[0091] Calculate the sum of counts corresponding to each category label within the preset number of times. If the sum of counts is greater than 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to 1. If the sum of counts is equal to 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to 0. If the sum of counts is less than 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to -1.

[0092] In this embodiment of the invention, the frequency value and trend change coefficient of each type of dynamic behavior data can reflect the changes in each type of dynamic behavior data during the observation time span, thereby reflecting the changes in user focus and degree of attention.

[0093] S4. Based on the classification labels, frequency values, and trend change coefficients, convert the corresponding dynamic behavior data into a dynamic vector matrix;

[0094] In this embodiment of the invention, the classification label, frequency value, and trend change coefficient corresponding to each type of dynamic behavior data are used as the three dimensions of data features of that type of dynamic behavior data. Based on these three dimensions of data features, the corresponding dynamic behavior data is converted into a dynamic vector matrix for subsequent calculations.

[0095] Specifically, S4 includes:

[0096] The category label, frequency value and trend change coefficient of the dynamic behavior data under the same category label are successively vectorized and concatenated to obtain the one-dimensional vector corresponding to the dynamic behavior data.

[0097] Obtain the dimension of the static vector matrix, and combine the one-dimensional vectors corresponding to all dynamic behavior data according to the dimension of the static vector matrix to obtain the dynamic vector matrix.

[0098] In this embodiment of the invention, the alignment of the dynamic vector matrix and the static vector matrix can be achieved using a matrix completion algorithm.

[0099] S5. The static vector matrix and the dynamic vector matrix are weighted using a preset attention mechanism, and the weighted static vector matrix and the weighted dynamic vector matrix are fused to obtain a fused vector matrix. The fused vector matrix is ​​then used to generate a recommendation result for the target user based on the pre-trained recommendation model.

[0100] In this embodiment of the invention, a pre-trained activation function can be used to weight the static vector matrix and the dynamic vector matrix. The pre-trained activation function can be a softmax activation function, a sigmoid activation function, a ReLU activation function, etc. The pre-trained activation function allows for dynamic weighting of both the static and dynamic vector matrices.

[0101] In this embodiment of the invention, the activation function can be trained by changing the ratio and quality of static data samples composed of static basic information and dynamic data samples composed of dynamic behavioral data, so as to ensure that the finally trained activation function can accurately and dynamically assign different weights to the static vector matrix and the dynamic vector matrix.

[0102] It is understandable that the information value embodied by the static vector matrix and the dynamic vector matrix of the same object is different. By assigning different weights to the static vector matrix and the dynamic vector matrix, more valuable information can be highlighted so that the pre-trained recommendation model can accurately predict the content that the target user is interested in.

[0103] In this embodiment of the invention, the pre-trained recommendation model may be a deep learning recommendation model based on neural networks.

[0104] This invention application starts from two dimensions: static basic information and dynamic behavioral data of users. On the one hand, by calculating the frequency value and trend change coefficient of each type of dynamic behavioral data, it reflects both the differences between user dynamic behavioral data and the essence of dynamic behavioral data changes. On the other hand, by using a preset attention mechanism to weight the static vector matrix and the dynamic vector matrix respectively, the weight of static basic data and dynamic behavioral data is dynamically determined, ensuring that the final fused vector matrix can truly reflect the changing trend of user interests, making the recommendation results closer to the user's current actual needs and improving the accuracy of the recommendation.

[0105] like Figure 3 This is a functional block diagram of a recommendation generation device based on static and dynamic data provided in an embodiment of the present invention.

[0106] The recommendation generation device 100 based on static and dynamic data described in this invention can be installed in an electronic device. Depending on the functions implemented, the recommendation generation device 100 based on static and dynamic data includes a raw data acquisition module 101, a static vector conversion module 102, a dynamic vector conversion module 103, a static-dynamic fusion module 104, and a recommendation prediction module 105. The modules described in this invention can also be called units, referring to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0107] In this embodiment, the functions of each module / unit are as follows:

[0108] The original data acquisition module 101 is used to acquire the static basic information and dynamic behavior data of the target user;

[0109] The static vector conversion module 102 is used to convert the static basic information into a static vector matrix.

[0110] The dynamic vector conversion module 103 is used to classify the dynamic behavior data, determine the classification label corresponding to the dynamic behavior data, calculate the frequency value and trend change coefficient of the dynamic behavior data under each classification label, and convert the corresponding dynamic behavior data into a dynamic vector matrix based on the classification label, frequency value and trend change coefficient.

[0111] The static-dynamic fusion module 104 is used to weight the static vector matrix and the dynamic vector matrix respectively using a preset attention mechanism, and fuse the weighted static vector matrix and the weighted dynamic vector matrix to obtain a fused vector matrix.

[0112] The recommendation prediction module 105 is used to generate a recommendation result to the target user based on the fusion vector matrix using a pre-trained recommendation model.

[0113] In detail, the specific implementation methods of each module of the recommendation generation device 100 based on static and dynamic data are as follows:

[0114] Step 1: Obtain the target user's static basic information and dynamic behavior data;

[0115] In this embodiment of the invention, the static basic information refers to the background information of the target user. This information has a certain degree of stability and is not easily changed in a short period of time. For example, the user's gender, education level, occupation, income level, city of residence, physical health status, and favorite sports.

[0116] In this embodiment of the invention, the dynamic behavioral data refers to a series of behavioral data of the target user within a specified observation period, surrounding the actual recommendation business scenario. This dynamic behavioral data has a certain degree of randomness and is prone to jumps. For example, in a recommendation scenario for insurance services, information such as the user's insurance business transactions or browsing behavior over the past two years can be collected, including purchasing a certain type of insurance, canceling a certain type of insurance, online consultation on a certain type of insurance, browsing insurance products, and complaints / reviews. Information such as medical treatment and property purchases by the user over the past two years can also be collected. It should be noted that all the information collected above is information that can be collected only after authorization from the target user.

[0117] It is understandable that the financial information that users pay attention to is closely related to their static basic information. For example, the higher a user's education level, the wider the range of financial products they may be interested in; similarly, residents of developed cities are more likely to engage in financial management than residents of ordinary cities. On the other hand, users' past dynamic behavior data can also reflect their interests. For instance, if a user frequently browses or inquires about a certain type of financial product within a short period of time, it indicates that the user is currently paying attention to that financial product. Therefore, this embodiment of the invention analyzes the content that a user may be interested in from both the user's static basic information and dynamic behavior data, ensuring the accuracy of the recommendation analysis source.

[0118] Step 2: Convert the static basic information into a static vector matrix;

[0119] In this embodiment of the invention, by converting the static basic information into a static vector matrix, the static basic information, which is text data composed of natural language, can be converted into vector data that can be recognized by the subsequent recommendation model based on neural network deep learning.

[0120] Specifically, the conversion of the static basic information into a static vector matrix includes:

[0121] The static basic information is semantically divided to obtain one or more static information units;

[0122] Extract keywords from each static information unit, and perform fuzzy matching between the keywords and preset static information tags to obtain the static information tags corresponding to each static information unit;

[0123] Obtain the value range corresponding to each static information tag, and convert the corresponding static information unit into a static numerical point using the segmented conversion method according to the value range.

[0124] Perform vector transformation operations on the static numerical points corresponding to all static information units to obtain the static vector matrix.

[0125] In this embodiment of the invention, an intelligent model with semantic recognition function can be used to perform semantic analysis on the static basic information, and the static basic information expressing the same semantic information can be divided into a static information unit.

[0126] In this embodiment of the invention, the preset static information tag refers to the information type used to describe the type of a certain static basic information. It can be set according to the actual business situation. For example, the preset static information tag includes, but is not limited to, tags such as gender, education level, income level, place of residence, and occupation.

[0127] In this embodiment of the invention, the value range corresponding to each static information tag can be set according to the actual situation. For example, for the static information tag of gender, the value range of the static information tag includes 1 or 0, where 1 represents male and 0 represents female.

[0128] In this embodiment of the invention, a segmented conversion method is used to convert the corresponding static information units into static numerical points. For example, if the target user has a high school education, the value range corresponding to the education level is [1-4], where 1 represents high school or below, 2 represents undergraduate, 3 represents postgraduate, and 4 represents postgraduate or above. Therefore, the static numerical point corresponding to the static basic data of the target user's education level static information tag is 1.

[0129] Step 3: Classify the dynamic behavior data, determine the corresponding classification labels for the dynamic behavior data, and calculate the frequency value and trend change coefficient of the dynamic behavior data under each classification label;

[0130] In this embodiment of the invention, the dynamic behavior of the target user can be classified according to actual business needs. For example, it can be distinguished by the nature of the action, including but not limited to browsing, consultation, assistance, and evaluation. It can also be classified according to the different objects involved in the same action, for example, based on the different target objects browsed, the dynamic behavior data includes but is not limited to fund browsing behavior, bond browsing behavior, and insurance browsing behavior. Furthermore, it can be distinguished according to the different means by which the action is performed, the dynamic behavior data can be classified as online consultation, telephone consultation, and on-site consultation, etc.

[0131] Understandably, the purpose of categorizing the dynamic behavioral data is to reflect the differences in user focus and level of attention among different types of dynamic behavioral data. For example, the level of user interest in an insurance product reflected by two types of behavioral data—one being a face-to-face consultation with staff at an insurance branch and the other being an online consultation with a smart customer service representative via an app—is different. Generally, face-to-face communication reflects a higher level of user interest in the product.

[0132] In this embodiment of the invention, clustering and other techniques can be used to classify the dynamic behavioral data.

[0133] In another optional embodiment of the present invention, the dynamic behavior data can be classified and the corresponding classification labels can be determined by the following method:

[0134] Extract the data features of each piece of dynamic behavior data and the tag features of each preset dynamic behavior tag in sequence;

[0135] The similarity between the data features of each piece of dynamic behavior data and each of the label features is calculated sequentially.

[0136] The dynamic behavior label corresponding to the maximum similarity is selected as the classification label of the corresponding dynamic behavior data.

[0137] In this embodiment of the invention, the preset dynamic behavior tags can be set according to actual business needs, such as online consultation tags, telephone consultation tags, and on-site consultation tags.

[0138] In this embodiment of the invention, a pre-trained machine learning model with semantic recognition function can be used to extract data features of each dynamic behavior data and label features of preset dynamic behavior tags. The data features include, but are not limited to, dynamic behavior occurrence path features, such as telephone or on-site; dynamic behavior target object features, such as insurance products, fund products, etc.; and dynamic behavior emotional features, such as evaluation terms like "very good" or "dissatisfied".

[0139] In this embodiment of the invention, the tag feature refers to the text feature of each preset dynamic behavior tag.

[0140] In this embodiment of the invention, the cosine similarity algorithm can be used to calculate the similarity between the data features and the label features.

[0141] In this embodiment of the invention, the frequency value of each type of dynamic behavior data refers to the total number of times that type of dynamic behavior occurs within the observation period, reflecting the overall activity level of that type of dynamic behavior data. The trend change coefficient of each type of dynamic behavior data refers to the activity trend exhibited by that type of dynamic behavior within the observation period, including becoming increasingly active, becoming increasingly inactive, and basically stable. For increasingly active behavior, the corresponding trend change coefficient is 1; for increasingly inactive behavior, the corresponding trend change coefficient is -1; and for basically stable behavior, the corresponding trend change coefficient is 0.

[0142] In this embodiment of the invention, the frequency value and trend change coefficient of dynamic behavior data under each category label can be calculated by the following method, including:

[0143] Using dynamic behavior data under the same category label as the target object, obtain the timestamp set of the target object;

[0144] From the set of timestamps, remove the smallest timestamp and randomly select two timestamps a preset number of times as the measurement point timestamps;

[0145] Calculate the difference in frequency values ​​of dynamic behavior data under the same category label for each of the two timestamps of the measurement points relative to the minimum timestamp;

[0146] When the difference between the frequency values ​​is greater than 0, the count is 1; when the difference between the frequency values ​​is less than 0, the count is -1; when the difference between the frequency values ​​is equal to 0, the count is 0.

[0147] Calculate the sum of counts corresponding to each category label within the preset number of times. If the sum of counts is greater than 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to 1. If the sum of counts is equal to 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to 0. If the sum of counts is less than 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to -1.

[0148] In this embodiment of the invention, the frequency value and trend change coefficient of each type of dynamic behavior data can reflect the changes in each type of dynamic behavior data during the observation time span, thereby reflecting the changes in user focus and degree of attention.

[0149] Step 4: Based on the classification labels, frequency values, and trend change coefficients, convert the corresponding dynamic behavior data into a dynamic vector matrix;

[0150] In this embodiment of the invention, the classification label, frequency value, and trend change coefficient corresponding to each type of dynamic behavior data are used as the three dimensions of data features of that type of dynamic behavior data. Based on these three dimensions of data features, the corresponding dynamic behavior data is converted into a dynamic vector matrix for subsequent calculations.

[0151] Specifically, the step of converting the corresponding dynamic behavior data into a dynamic vector matrix based on the classification label, frequency value, and trend change coefficient includes:

[0152] The category label, frequency value and trend change coefficient of the dynamic behavior data under the same category label are successively vectorized and concatenated to obtain the one-dimensional vector corresponding to the dynamic behavior data.

[0153] Obtain the dimension of the static vector matrix, and combine the one-dimensional vectors corresponding to all dynamic behavior data according to the dimension of the static vector matrix to obtain the dynamic vector matrix.

[0154] In this embodiment of the invention, the alignment of the dynamic vector matrix and the static vector matrix can be achieved using a matrix completion algorithm.

[0155] Step 5: Use a preset attention mechanism to weight the static vector matrix and the dynamic vector matrix respectively, fuse the weighted static vector matrix and the weighted dynamic vector matrix to obtain a fused vector matrix, and use a pre-trained recommendation model to generate a recommendation result to the target user based on the fused vector matrix.

[0156] In this embodiment of the invention, a pre-trained activation function can be used to weight the static vector matrix and the dynamic vector matrix. The pre-trained activation function can be a softmax activation function, a sigmoid activation function, a ReLU activation function, etc. The pre-trained activation function allows for dynamic weighting of both the static and dynamic vector matrices.

[0157] In this embodiment of the invention, the activation function can be trained by changing the ratio and quality of static data samples composed of static basic information and dynamic data samples composed of dynamic behavioral data, so as to ensure that the finally trained activation function can accurately and dynamically assign different weights to the static vector matrix and the dynamic vector matrix.

[0158] It is understandable that the information value embodied by the static vector matrix and the dynamic vector matrix of the same object is different. By assigning different weights to the static vector matrix and the dynamic vector matrix, more valuable information can be highlighted so that the pre-trained recommendation model can accurately predict the content that the target user is interested in.

[0159] In this embodiment of the invention, the pre-trained recommendation model may be a deep learning recommendation model based on neural networks.

[0160] This invention application starts from two dimensions: static basic information and dynamic behavioral data of users. On the one hand, by calculating the frequency value and trend change coefficient of each type of dynamic behavioral data, it reflects both the differences between user dynamic behavioral data and the essence of dynamic behavioral data changes. On the other hand, by using a preset attention mechanism to weight the static vector matrix and the dynamic vector matrix respectively, the weight of static basic data and dynamic behavioral data is dynamically determined, ensuring that the final fused vector matrix can truly reflect the changing trend of user interests, making the recommendation results closer to the user's current actual needs and improving the accuracy of the recommendation.

[0161] like Figure 4 The diagram shown is a structural schematic of an electronic device that implements a recommendation generation method based on static and dynamic data, according to an embodiment of the present invention.

[0162] The electronic device 1 may include a processor 10, a memory 11, and a bus, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a recommendation based on static and dynamic data.

[0163] The memory 11 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a portable hard drive. In other embodiments, the memory 11 can be an external storage device of the electronic device 1, such as a plug-in portable hard drive, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the electronic device 1. Furthermore, the memory 11 can include both internal and external storage units of the electronic device 1. The memory 11 can be used not only to store application software and various types of data installed on the electronic device 1, such as recommended code based on static and dynamic data, but also to temporarily store data that has been output or will be output.

[0164] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., recommendations based on static and dynamic data) and calls data stored in the memory 11 to perform various functions of the electronic device 1 and process data.

[0165] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable communication between the memory 11 and at least one processor 10, etc.

[0166] Figure 4 Only electronic devices with components are shown; those skilled in the art will understand that... Figure 4The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0167] For example, although not shown, the electronic device 1 may also include a power supply (such as a battery) to power various components. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0168] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is typically used to establish communication connections between the electronic device 1 and other electronic devices.

[0169] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired or wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device 1 and to display a visual user interface.

[0170] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0171] The recommendations stored in the memory 11 of the electronic device 1, based on static and dynamic data, are combinations of multiple instructions that, when run in the processor 10, can achieve the following:

[0172] Obtain static basic information and dynamic behavioral data of the target user;

[0173] Convert the static basic information into a static vector matrix;

[0174] The dynamic behavior data is classified to determine the corresponding classification labels, and the frequency value and trend change coefficient of the dynamic behavior data under each classification label are calculated.

[0175] Based on the classification labels, frequency values, and trend change coefficients, the corresponding dynamic behavior data is converted into a dynamic vector matrix;

[0176] The static vector matrix and the dynamic vector matrix are weighted separately using a preset attention mechanism, and the weighted static vector matrix and the weighted dynamic vector matrix are fused to obtain a fused vector matrix.

[0177] Using a pre-trained recommendation model, recommendations are generated for the target user based on the fusion vector matrix.

[0178] Furthermore, if the modules / units integrated in the electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0179] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following:

[0180] Obtain static basic information and dynamic behavioral data of the target user;

[0181] Convert the static basic information into a static vector matrix;

[0182] The dynamic behavior data is classified to determine the corresponding classification labels, and the frequency value and trend change coefficient of the dynamic behavior data under each classification label are calculated.

[0183] Based on the classification labels, frequency values, and trend change coefficients, the corresponding dynamic behavior data is converted into a dynamic vector matrix;

[0184] The static vector matrix and the dynamic vector matrix are weighted separately using a preset attention mechanism, and the weighted static vector matrix and the weighted dynamic vector matrix are fused to obtain a fused vector matrix.

[0185] Using a pre-trained recommendation model, recommendations are generated for the target user based on the fusion vector matrix.

[0186] Furthermore, the functional modules in the various embodiments of the present invention 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 integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0187] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0188] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0189] The blockchain referred to in this invention is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0190] The embodiments of this application can acquire and process relevant data based on holographic projection technology. Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0191] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. The term "second class" is used to indicate names and does not indicate any specific order.

[0192] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A recommendation generation method based on static and dynamic data, characterized in that, The method includes: Obtain static basic information and dynamic behavioral data of the target user; Convert the static basic information into a static vector matrix; The data features of each dynamic behavior data and the tag features of each preset dynamic behavior tag are extracted sequentially. The similarity between the data features of each dynamic behavior data and each tag feature is calculated successively. The dynamic behavior tag corresponding to the maximum similarity is selected as the classification tag of the corresponding dynamic behavior data. The frequency value and trend change coefficient of the dynamic behavior data under each classification tag are calculated. Based on the classification labels, frequency values, and trend change coefficients, the corresponding dynamic behavior data is converted into a dynamic vector matrix; The static vector matrix and the dynamic vector matrix are weighted separately using a preset attention mechanism, and the weighted static vector matrix and the weighted dynamic vector matrix are fused to obtain a fused vector matrix. Using a pre-trained recommendation model, a recommendation result is generated for the target user based on the fusion vector matrix; The calculation of the frequency value and trend change coefficient of dynamic behavior data under each category label includes: Using dynamic behavior data under the same category label as the target object, obtain the timestamp set of the target object; From the set of timestamps, remove the smallest timestamp and randomly select two timestamps a preset number of times as the measurement point timestamps; Calculate the difference in frequency values ​​of dynamic behavior data under the same category label for each of the two timestamps of the measurement points relative to the minimum timestamp; When the difference between the frequency values ​​is greater than 0, the count is 1; when the difference between the frequency values ​​is less than 0, the count is -1; when the difference between the frequency values ​​is equal to 0, the count is 0. Calculate the sum of counts corresponding to each category label within the preset number of times. If the sum of counts is greater than 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to 1. If the sum of counts is equal to 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to 0. If the sum of counts is less than 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to -1.

2. The recommendation generation method based on static and dynamic data as described in claim 1, characterized in that, The step of converting the static basic information into a static vector matrix includes: The static basic information is semantically divided to obtain one or more static information units; Extract keywords from each static information unit, and perform fuzzy matching between the keywords and preset static information tags to obtain the static information tags corresponding to each static information unit; Obtain the value range corresponding to each static information tag, and convert the corresponding static information unit into a static numerical point using the segmented conversion method according to the value range. Perform vector transformation operations on the static numerical points corresponding to all static information units to obtain the static vector matrix.

3. The recommendation generation method based on static and dynamic data as described in claim 1, characterized in that, The step of converting the corresponding dynamic behavior data into a dynamic vector matrix based on the classification label, frequency value, and trend change coefficient includes: The category label, frequency value and trend change coefficient of the dynamic behavior data under the same category label are successively vectorized and concatenated to obtain the one-dimensional vector corresponding to the dynamic behavior data. Obtain the dimension of the static vector matrix, and combine the one-dimensional vectors corresponding to all dynamic behavior data according to the dimension of the static vector matrix to obtain the dynamic vector matrix.

4. A recommendation generation device based on static and dynamic data, characterized in that, The device includes: The original data acquisition module is used to acquire the static basic information and dynamic behavior data of the target user; A static vector conversion module is used to convert the static basic information into a static vector matrix; The dynamic vector conversion module is used to sequentially extract the data features of each dynamic behavior data and the label features of each preset dynamic behavior label, calculate the similarity between the data features of each dynamic behavior data and each label feature, select the dynamic behavior label corresponding to the maximum similarity as the classification label of the corresponding dynamic behavior data, calculate the frequency value and trend change coefficient of the dynamic behavior data under each classification label, and convert the corresponding dynamic behavior data into a dynamic vector matrix based on the classification label, frequency value and trend change coefficient. The static and dynamic fusion module is used to weight the static vector matrix and the dynamic vector matrix respectively using a preset attention mechanism, and then fuse the weighted static vector matrix and the weighted dynamic vector matrix to obtain a fused vector matrix. The recommendation prediction module is used to generate recommendations to the target user based on the fusion vector matrix using a pre-trained recommendation model. The calculation of the frequency value and trend change coefficient of dynamic behavior data under each category label includes: Using dynamic behavior data under the same category label as the target object, obtain the timestamp set of the target object; From the set of timestamps, remove the smallest timestamp and randomly select two timestamps a preset number of times as the measurement point timestamps; Calculate the difference in frequency values ​​of dynamic behavior data under the same category label for each of the two timestamps of the measurement points relative to the minimum timestamp; When the difference between the frequency values ​​is greater than 0, the count is 1; when the difference between the frequency values ​​is less than 0, the count is -1; when the difference between the frequency values ​​is equal to 0, the count is 0. Calculate the sum of counts corresponding to each category label within the preset number of times. If the sum of counts is greater than 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to 1. If the sum of counts is equal to 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to 0. If the sum of counts is less than 0, set the trend change coefficient of the dynamic behavior data under the corresponding category label to -1.

5. The recommendation generation device based on static and dynamic data as described in claim 4, characterized in that, The original data acquisition module converts the static basic information into a static vector matrix using the following method: The static basic information is semantically divided to obtain one or more static information units; Extract keywords from each static information unit, and perform fuzzy matching between the keywords and preset static information tags to obtain the static information tags corresponding to each static information unit; Obtain the value range corresponding to each static information tag, and convert the corresponding static information unit into a static numerical point using the segmented conversion method according to the value range. Perform vector transformation operations on the static numerical points corresponding to all static information units to obtain the static vector matrix.

6. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor to enable the at least one processor to perform the recommendation generation method based on static and dynamic data as described in any one of claims 1 to 3.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the recommendation generation method based on static and dynamic data as described in any one of claims 1 to 3.

Images