CN112990323A

CN112990323A - User portrait mining method based on big data online mode and machine learning system

Info

Publication number: CN112990323A
Application number: CN202110308304.4A
Authority: CN
Inventors: 李光伟
Original assignee: Individual
Current assignee: Individual
Priority date: 2021-03-23
Filing date: 2021-03-23
Publication date: 2021-06-18

Abstract

The embodiment of the disclosure provides a user portrait mining method and a machine learning system based on a big data line mode, and the accuracy of user portrait mining is improved by integrating data evaluation schemes of various service interaction processes, wherein if a service interaction index is used for analyzing the service interaction process based on interest characteristics of information interaction behaviors, and service coverage information is used for analyzing the service interaction process through service use characteristics of users.

Description

User portrait mining method based on big data online mode and machine learning system

Technical Field

The present disclosure relates to the field of big data technologies, and in particular, to a user portrait mining method and a machine learning system based on a big data online mode.

Background

The user representation refers to a tagged user model abstracted according to information such as user attributes, user preferences, living habits, user behaviors and the like. Colloquially, a user is labeled, and the label is a highly refined characteristic mark obtained by analyzing user information. By tagging, a user may be described with some highly generalized, easily understandable features that may make it easier for a person to understand the user and may facilitate computer processing. In the related art, how to improve the precision of user portrait mining is a technical problem to be considered and solved urgently.

Disclosure of Invention

In order to overcome at least the above-mentioned deficiencies in the prior art, the present disclosure provides a method for mining a user portrait based on a large online pattern and a machine learning system.

In a first aspect, the present disclosure provides a user portrait mining method based on a big data online mode, which is applied to a machine learning system, where the machine learning system is communicatively connected to a plurality of business service devices, and the method includes:

acquiring an information interaction behavior sequence corresponding to each service interaction process according to big data information service of a service user of the service equipment, wherein the information interaction behavior sequence comprises interaction behavior data of at least two target interaction information, and the interaction behavior data comprises information interaction behavior of the target interaction information when a target micro service is called each time and a calling service node;

determining a business interaction index of each information interaction behavior sequence according to information interaction behaviors when each target interaction information calls the target micro service, wherein the business interaction index is used for representing business interestingness of the target interaction information in the information interaction behavior sequence when the target micro service is called;

determining service coverage information of each information interaction behavior sequence according to a calling service node when each target interaction information calls the target micro service each time; the service coverage information is used for representing the service coverage degree of the target micro service called by the target interaction information in the information interaction behavior sequence;

and mining the user portrait in the service interaction process according to the service interaction indexes and the service coverage information of the information interaction behavior sequence corresponding to each service interaction process.

In a second aspect, an embodiment of the present disclosure further provides a user portrait mining system based on a big data online mode, where the user portrait mining system based on the big data online mode includes a machine learning system and a plurality of business service devices communicatively connected to the machine learning system;

the machine learning system to:

According to any one of the aspects, the data evaluation scheme is used for integrating the data evaluation schemes of various service interaction processes, if the service interaction indexes are used for analyzing the service interaction processes based on the interest characteristics of the information interaction behaviors, the service coverage information is used for analyzing the service interaction processes through the service use characteristics of the users, and the interest characteristics and the service use characteristics of the information interaction behaviors can be simultaneously integrated by combining the service use characteristics and the service use characteristics, so that the accuracy of user portrait mining is improved.

Drawings

FIG. 1 is a schematic diagram of an application scenario of a user portrait mining system based on a big data online mode according to an embodiment of the present disclosure;

FIG. 2 is a schematic flow chart illustrating a user portrait mining method based on a big data online mode according to an embodiment of the present disclosure;

FIG. 3 is a block diagram illustrating a user portrait mining apparatus based on a big data line mode according to an embodiment of the present disclosure;

FIG. 4 is a block diagram illustrating the structural components of a machine learning system for implementing the above-described user portrait mining method based on patterns on a big data line according to an embodiment of the present disclosure.

Detailed Description

The present disclosure is described in detail below with reference to the drawings, and the specific operation methods in the method embodiments can also be applied to the device embodiments or the system embodiments.

FIG. 1 is an interaction diagram of a large data online schema-based user representation mining system 10, provided by an embodiment of the present disclosure. The user representation mining system 10 based on large data online patterns may include a machine learning system 100 and a business service appliance 200 communicatively connected to the machine learning system 100. While the user representation mining system 10 based on a big data-line schema shown in FIG. 1 is merely one possible example, in other possible embodiments, the user representation mining system 10 based on a big data-line schema may include only at least some of the components shown in FIG. 1 or may include additional components.

In a separate embodiment, the machine learning system 100 and the business service apparatus 200 in the big data line mode-based user representation mining system 10 may cooperatively perform the big data line mode-based user representation mining method described in the following method embodiments, and the detailed description of the method embodiments may be referred to in the detailed description of the following steps of the machine learning system 100 and the business service apparatus 200.

Based on this, in order to solve the technical problem in the foregoing background, fig. 2 is a flowchart illustrating a user portrait mining method based on a pattern on a big data line according to an embodiment of the present disclosure, which may be executed by the machine learning system 100 shown in fig. 1, and the user portrait mining method based on a pattern on a big data line is described in detail below.

Step S110, obtaining an information interaction behavior sequence corresponding to each service interaction process according to the big data information service of the service user of the service equipment 200.

The information interaction behavior sequence can comprise interaction behavior data of at least two target interaction information, and the interaction behavior data comprises information interaction behaviors of the target interaction information when the target micro service is called every time and calling service nodes. The interactive information may refer to an information page with a dynamic link, and may include text content, audio/video content, service link content, and the like of the information, and the information interaction behavior may refer to information interaction in the page of the interactive information by the user, including but not limited to a praise behavior, a cancel behavior, and the like. Invoking a service node may refer to, but is not limited to, a point in time, or a node of a logical service interaction.

In a separate embodiment, step S110 may be implemented by the following steps.

Step S111, obtaining the interaction behavior data of each target interaction information. The interactive behavior data comprises a business interactive process ID and information interactive behavior and calling business nodes when the target micro service is called by the target interactive information each time, and the business interactive process ID is used for indicating the target interactive information to obtain a business interactive process of the target micro service.

In this embodiment, a micro service which is currently and urgently to be promoted may be used as a target micro service, interaction information configured with the target micro service is used as target interaction information, and an interaction behavior acquisition program is added in a configuration packet of the target micro service, so that, when the target interaction information calls the target micro service each time, the interaction behavior acquisition program can acquire an information interaction behavior and a calling service node of the current target interaction information, and report the acquired information interaction behavior and calling service node to a server, and report to the server an information identifier and a service interaction process ID of the target interaction information, where the information identifier is used to uniquely identify the target interaction information, the service interaction process ID is used to indicate a source of the target micro service in the target interaction information, and after receiving information reported by the interaction behavior acquisition program, the server identifies the information interaction behavior, the interaction behavior, and the calling service belonging to the same target interaction information according to the information identifier, And summarizing the calling service node, the service interaction process ID and the information identification to the corresponding target interaction information name to obtain interaction behavior data of each target interaction information, wherein the interaction behavior data can be regarded as a data set, and each piece of data in the data set corresponds to one calling action of the target interaction information to the target micro-service. For example, the target micro service calls the target interaction information A n (n ≧ 1 integer) times, and the interaction behavior data of the target interaction information A includes n pieces of data.

The information interaction behaviors obtained by the interaction behavior obtaining program can specifically include a plurality of information interaction behaviors, except that the reported information id is used for identifying information and the service interaction process id is used for identifying a service interaction process, the main characteristics are divided into the following categories, namely military categories, scientific categories, short video categories and the like.

Step S112, classifying the interaction behavior data of each target interaction information according to the service interaction process ID to obtain candidate information interaction behavior sequences corresponding to the service interaction processes indicated by each service interaction process ID, wherein each candidate information interaction behavior sequence comprises interaction behavior data of at least two target interaction information, and all target interaction information in the same candidate information interaction behavior sequence have the same service interaction process ID.

After the interaction behavior data of each target interaction information is obtained, the interaction behavior data can be classified according to the service interaction process ID of the interaction behavior data, namely, the interaction behavior data with the same service interaction process ID are classified into the same candidate information interaction behavior sequence, so that the service interaction process ID of each candidate information interaction behavior sequence is different, and different candidate information interaction behavior sequences correspond to different service interaction processes because the service interaction process ID indicates the source service interaction process of the target micro-service. The step classifies the interaction behavior data of the target interaction information according to the service interaction process, so that the subsequent analysis of the detail characteristics of each service interaction process based on the interaction behavior characteristics of the target interaction information is facilitated.

Since if the amount of information in a certain service interaction process is too small, the service interaction index may have a large error, the service interaction process with too small amount of information needs to be filtered out. Therefore, after step S112, step S113 may also be included: and counting the data quantity of the interaction behavior data in each candidate information interaction behavior sequence, and taking the candidate information interaction behavior sequence with the data quantity more than or equal to a preset threshold value as the information interaction behavior sequence. The candidate information interaction behavior sequence with the data volume smaller than the preset threshold value is filtered to obtain the information interaction behavior sequence used for analyzing the value of the service interaction process, and the accuracy of calculation of subsequent service interaction indexes can be improved. The preset threshold value is typically hundreds to thousands.

Step S120, determining a business interaction index of each information interaction behavior sequence according to the information interaction behavior when each target interaction information calls the target micro service, wherein the business interaction index is used for representing the business interest degree of the target interaction information in the information interaction behavior sequence when the target micro service is called.

The service interaction indexes of each information interaction behavior sequence comprise a first service interaction index and a second service interaction index. In an independent embodiment, step S120 may specifically include: determining a first service interaction index of each information interaction behavior sequence according to the information interaction behavior of each target interaction information when the target micro service is called for the first time; the first business interaction index is used for representing the business interest degree of target interaction information in the information interaction behavior sequence when the target micro service is configured; determining a second service interaction index of each information interaction behavior sequence according to the information interaction behavior of each target interaction information when the target micro service is called for the first time in each interaction stage; the second business interaction index is used for representing the business interest degree of the target interaction information in the information interaction behavior sequence when the target micro service is started; and using the first service interaction index and the second service interaction index of each information interaction behavior sequence as service interaction indexes of the information interaction behavior sequence.

In an independent embodiment, the business interest detection for each business interaction process based on a single information interaction behavior may include the following steps.

Step S201, the information interaction behavior when each target interaction information calls the target micro service for the first time is used as the first information interaction behavior of the target interaction information.

The information interaction behavior when the target interaction information calls the target micro service for the first time is the information interaction behavior when the target interaction information configures the target micro service. For example, for the interaction behavior data of any target interaction information, the earliest occurring information interaction behavior can be found according to the calling service node, and is used as the information interaction behavior when the target interaction information calls the target micro service for the first time.

Here, only the information interaction behavior when the target interaction information calls the target micro-service for the first time is obtained because the interaction information with higher business interest in the information interaction behavior in all the target interaction information needs to be searched, and therefore the influence factor of each target interaction information needs to be the same. If the interaction behavior data of the target interaction information is used each time the target micro service is started, the influence factors of each target interaction information are different and are not comparable.

Step S202, the first information interaction behavior of each target interaction information is processed by feature processing, and first interaction behavior features corresponding to each target interaction information are obtained.

After the first information interaction behavior corresponding to each target interaction information is obtained, which features are needed to be used for service interest detection is selected. Because the assumption that the more business interest the behavior characteristics of information interaction is, the more possible information is the interest information needs to be satisfied, the selected characteristics need to satisfy the condition that the distribution is more consistent under normal conditions, different characteristic values and characteristic value combinations are presented under the condition of interest confirmation, and the ratio of the business interest characteristic value to the characteristic value combination is very small.

The characteristic processing of the first information interaction behavior comprises the following steps: extracting features for business interest analysis from the first information interaction behavior, and processing the extracted features into feature vectors that can be used for calculation.

In a separate embodiment, all kinds of information interaction behavior features included in the first information interaction behavior can be classified into two categories, dynamic features and static features. Wherein the dynamic characteristics refer to characteristics changed along with the change of the interaction behavior, and are mainly characteristics of target interaction information which can be changed at any time; static features are features that cannot be easily changed by the user, mainly some features of the target interaction information itself. And taking the dynamic characteristics and the static characteristics as the characteristics for business interest analysis. Further, processing the dynamic features and the static features into feature vectors which can be directly input by the model, mainly converting the features into character strings and shaping numerical values, wherein the shaping of the numerical values means that: and according to the range in which the numerical value falls, taking the designated character corresponding to the range as the shaping result of the numerical value.

Step S203, inputting the first interaction behavior characteristics of each target interaction information into a service interest prediction network, and performing service interest characteristic analysis on the first interaction behavior characteristics of each target interaction information based on the service interest prediction network to obtain a first service interest value of each target interaction information.

After the first interaction behavior characteristics of the target interaction information are obtained, the target interaction information of the network detection service interest can be predicted by adopting the service interest. The business interest detection algorithm may be any One of an isolated Forest algorithm (Isolation Forest algorithm), a class vector machine (One-ClassSVM), a Gaussian probability density-based business interest point detection (Elliptic Envelope), an ensemble learning method-based business interest point detection (Isolation Forest), a density-based Local business interest factor (Local outlying factor), and the like, or a plurality of business interest detection algorithms are adopted to vote or integrate a plurality of detection results, and the like.

In a separate embodiment, n pieces of annotation basic data can be randomly selected from the annotation information and input into an isolated tree. Randomly assigning a data dimension to the n samples and randomly generating a splitting point according to the dimension, wherein the splitting point divides the n samples into a left sub-tree and a right sub-tree. And (4) continuously randomly appointing one dimension for the rest data dimensions and dividing the sub-trees, and repeating the process until only one data on the leaf node can not be re-split, so as to obtain a single isolated tree. M isolated trees are obtained by using the same method, namely n samples are randomly selected as root nodes, and the tree is repeatedly built from the root nodes of the tree. These trees together constitute an isolated forest and the final result needs to be generated jointly by combining the results of all the trees. The service interest degree of the sample is determined by the path length from the root node to the leaf node, and since sparsely distributed service interest points can be divided by only splitting for a few times, the path lengths of the sparsely distributed service interest points are usually smaller. The business interest score of the final sample must be a business interest point if it is close to 1, and must not be a business interest point if it is much less than 0.5.

The isolated forest model mainly has 3 hyper-parameters which need to be preset, and the hyper-parameters are the maximum depth of the tree, the number of all trees in the forest and the number of samples when the tree is built each time. In the embodiment of the disclosure, the maximum depth of the tree may be set to be between 10 and 100, or is not limited directly; the number of all trees in the forest is 100- & lt200- & gt; the number of samples is 256, since too many samples may reduce the ability of the model to identify business interest data.

During model prediction, for input to-be-detected data, firstly traversing all trees to respectively obtain the path heights of the to-be-detected data reaching leaf nodes, then calculating the average value h (x) of all the heights, and then obtaining a sample business interest score value p (x, m), wherein the business interest probability is larger when the business interest score value is closer to 1.

And inputting the first interaction behavior characteristics of each target interaction information into an isolated forest model, and performing service interest characteristic analysis on the first interaction behavior characteristics of each target interaction information based on the isolated forest model to obtain a model service interest score, namely a first service interest value, of each target interaction information.

Step S204, based on the total amount of the target interaction information in each information interaction behavior sequence and the first business interest value of the target interaction information of which the first business interest value is greater than the first threshold value, determining a first business interaction index corresponding to each information interaction behavior sequence.

After obtaining the first service interest value of each target interaction information, target interaction information with the first service interest value larger than a first threshold (which may be 0.8) may be selected, and these target interaction information may be regarded as suspected service interest information. Calculating the total score of all suspected business interest information in each information interaction behavior sequence according to a first business interest value of the suspected business interest information, then dividing the total score by the total number of target interaction information in the information interaction behavior sequence to obtain the average business interest score of each information interaction behavior sequence, taking the average business interest score as a first business interaction index, wherein the higher the first business interaction index is, the higher the value of the business interaction process is.

In an independent embodiment, the business interest detection corresponding to the business interaction process based on the plurality of information interaction behaviors may include the following steps.

Step S301, using the information interaction behavior corresponding to the first calling of the target micro service in each interaction stage of each target interaction information as the second information interaction behavior.

Step S302, the target interaction information with the number of the second information interaction behaviors larger than or equal to the preset number threshold is determined as candidate interaction information.

The part is expected to mine service interest information by detecting the change condition of a plurality of information interaction behaviors corresponding to target micro-services before and after target interaction information logging, so that the data filtering is not only used for filtering the information interaction behaviors in the first configuration, but also used for filtering first information interaction behavior data in each period from the same target interaction information according to the period to serve as second information interaction behaviors, and then aggregating the second information interaction behavior data of the same target interaction information. That is, one or more second information interaction behaviors can be obtained for the same target interaction information. And filtering all the second information interaction behaviors of the same aggregated target interaction information to obtain target interaction information of which the data volume of the second information interaction behaviors is smaller than a preset number threshold value n, taking the target interaction information of which the number of the second information interaction behaviors is larger than or equal to n as candidate interaction information, wherein n is a defined threshold value and can be defined as 5-10, and the purpose is to reduce the influence of the characteristics of the information of which the number of the second information interaction behaviors is too small on the overall result.

Step S303, performing feature processing on all second information interaction behaviors of each candidate interaction information to obtain second interaction behavior features corresponding to each candidate interaction information.

After all the second information interaction behaviors of each candidate interaction information are obtained, feature processing needs to be performed on a plurality of second information interaction behaviors of each candidate interaction information, and second interaction behavior features corresponding to each candidate interaction information are generated. Wherein, the processing the characteristics of the plurality of second information interaction behaviors of the candidate interaction information comprises the following steps: extracting features for business interest analysis from the second information interaction behavior, and processing the extracted features into feature vectors that can be used for calculation. Since the part detects the change of the interaction behavior of a plurality of pieces of information, the adopted characteristics are basically dynamic characteristics, namely the characteristics change along with the behavior of the user. Further, the extracted features are processed into feature vectors which can be used for calculation, and mainly, changes of the features in a preset time are converted into characters or numerical values.

Step S304, inputting the second interaction behavior characteristics of each candidate interaction information into the service interest prediction network, and performing service interest analysis on the second interaction behavior characteristics of each candidate interaction information based on the service interest prediction network to obtain a second service interest value of each candidate interaction information.

After the second interaction behavior characteristics of the candidate interaction information in each information interaction behavior sequence are obtained, the target interaction information of the network detection service interest can be predicted by adopting the service interest. In the embodiment of the disclosure, an Isolation Forest (Isolation Forest) algorithm is adopted as the business interest prediction network. Please refer to the above description for the isolated forest algorithm, which is not described herein.

And inputting the second interaction behavior characteristics of each candidate interaction information into the isolated forest model, and performing business interest characteristic analysis on the second interaction behavior characteristics of each candidate interaction information based on the isolated forest model to obtain a model business interest score, namely a second business interest value, of each candidate interaction information.

Step S305, determining a second service interaction indicator corresponding to each information interaction behavior sequence based on the total number of candidate interaction information in each information interaction behavior sequence and a second service interest value of the candidate interaction information of which the second service interest value is greater than a second threshold.

After obtaining the second service interest value of each candidate interaction information, candidate interaction information having the second service interest value greater than the first threshold (which may be 0.8) may be selected, and these candidate interaction information may be regarded as suspected service interest information. And calculating the total score of all suspected business interest information in each information interaction behavior sequence according to a second business interest value of the suspected business interest information, then dividing the total score by the total number of candidate interaction information in the information interaction behavior sequence to obtain the average business interest score of each information interaction behavior sequence, taking the average business interest score as a second business interaction index, and obtaining the relatively worse value of the business interaction process when the second business interaction index is higher.

Step S130, determining the service coverage information of each information interaction behavior sequence according to the calling service node when each target interaction information calls the target micro service each time; the service coverage information is used for representing the service coverage degree of the target micro service called by the target interaction information in the information interaction behavior sequence.

In a separate embodiment, the method for calculating the service coverage information may include the following steps.

Step S401, according to the calling service node when each target interaction information calls the target micro service each time, counting the number of target interaction information for configuring the target micro service at the first calling service node and starting the target micro service at the second calling service node, the number of target interaction information for configuring the target micro service at the first calling service node, the number of target interaction information for starting the target micro service at the first calling service node and the second calling service node and the number of target interaction information for starting the target micro service at the first calling service node in each information interaction behavior sequence; the second calling service node is a calling service node behind the first calling service node.

Step S402, calculating the first service coverage information corresponding to each information interaction behavior sequence according to the quantity of the target interaction information for configuring the target micro service at the first calling service node and starting the target micro service at the second calling service node and the quantity of the target interaction information for configuring the target micro service at the first calling service node in each information interaction behavior sequence.

Step S403, calculating second service coverage information corresponding to each information interaction behavior sequence according to the number of target interaction information for enabling the target micro service at the first call service node and the second call service node in each information interaction behavior sequence and the number of target interaction information for enabling the target micro service at the first call service node.

Step S404, determining the service coverage information corresponding to each information interaction behavior sequence according to the first service coverage information and the second service coverage information corresponding to the information interaction behavior sequence.

The embodiment of the present disclosure mainly calculates two kinds of service coverage information, which are the service coverage information of the original interactive process and the service coverage information of the progress interactive process. The target interaction information of the target micro service configured at the first calling service node is regarded as an original interaction process, and the target interaction information configured before the first calling service node and enabling the target micro service at the first calling service node is regarded as a progress interaction process. The service coverage information of the original interaction process, i.e. the first service coverage information, refers to a ratio of the number of target interaction information for configuring the target micro service at the first calling service node and enabling the target micro service at the second calling service node in the information interaction behavior sequence to the number of target interaction information for configuring the target micro service at the first calling service node. The second service coverage information, which is the service coverage information of the progress interaction process, is a ratio of the number of target interaction information for enabling the target micro service at the first calling service node and the second calling service node in the information interaction behavior sequence to the number of target interaction information for enabling the target micro service at the first calling service node.

The following are exemplary: the service coverage information of the original interactive process may refer to a ratio of the amount of information that the target micro-service is configured and still activated on the nth day to the amount of information that the target micro-service is configured on the current day. The value of n is changed, and 1-day service coverage information, 3-day service coverage information, 1-week service coverage information, 2-week service coverage information, 3-week service coverage information and 1-month service coverage information of the interactive process can be calculated according to the service interactive process.

The service coverage information of the progress interactive process may refer to a ratio of the number of information that the target micro-service is started on the current day and the target micro-service is still started on the nth day to the number of information that the target micro-service is currently started. The value of n is changed, and 1-day service coverage information, 3-day service coverage information, 1-week service coverage information, 2-week service coverage information, 3-week service coverage information and 1-month service coverage information of the interactive process can be calculated according to the service interactive process.

And for each information interaction behavior sequence, calculating the average value of the first service coverage information and the second service coverage information of each information interaction behavior sequence, and taking the average value as the service coverage information corresponding to the information interaction behavior sequence.

Step S140, mining the user portrait in the service interaction process according to the service interaction index and the service coverage information of the information interaction behavior sequence corresponding to each service interaction process.

The service interaction indexes of the information interaction behavior sequence comprise: the method comprises a first service interaction index obtained based on single information interaction behavior service interest detection and a second service interaction index obtained based on multiple information interaction behavior service interest detection. In one possible implementation, the mining of the user portrait in the service interaction process according to the service interaction index and the service coverage information of the information interaction behavior sequence corresponding to each service interaction process may include the following steps

Step S141, the first service interaction index, the second service interaction index and the service coverage information of each information interaction behavior sequence are subjected to unified rule conversion processing, and a first unified rule value corresponding to the first service interaction index, a second unified rule value corresponding to the second service interaction index and a third unified rule value corresponding to the service coverage information are obtained.

Step S142, obtaining a first impact factor corresponding to the first service interaction indicator, a second impact factor corresponding to the second service interaction indicator, and a third impact factor corresponding to the service coverage information.

Step S143, according to the first influence factor, the second influence factor and the third influence factor, performing weighted calculation on the first uniform rule value, the second uniform rule value and the third uniform rule value of each information interaction behavior sequence to obtain a business interest value of each information interaction behavior sequence.

Step S144, mining the user portrait in the service interaction process according to the service interest value of the information interaction behavior sequence corresponding to each service interaction process.

In this embodiment, the numerical difference between the first service interaction index, the second service interaction index, and the service coverage information may be very large, and in order to integrate the scores of the service interaction processes obtained by the three methods to perform the order arrangement of the service interaction processes, the scores obtained by the three methods need to be subjected to a uniform rule conversion process, where the uniform rule conversion process refers to processing the first service interaction index, the second service interaction index, and the service coverage information in proportion so as to make them fall into a specific interval. Common unified rule conversion processing methods include: min-max normalization (Min-max normalization), log function transformation, atan function transformation, z-score normalization (zero-normalization), fuzzy quantization.

In one possible implementation manner, a unified rule transformation process may be performed on the first service interaction index, the second service interaction index and the service coverage information respectively by using a standardized formula x '— (x- μ)/σ, to obtain a first unified rule value corresponding to the first service interaction index, a second unified rule value corresponding to the second service interaction index and a third unified rule value corresponding to the service coverage information, where x is a score to be processed, and is the first service interaction index, the second service interaction index or the service coverage information, x' is a unified rule value corresponding to x, μ is a mean value of scores corresponding to the information interaction behavior sequences, σ is a standard deviation of scores corresponding to the information interaction behavior sequences, and μ is a mean value of the first service interaction index corresponding to the information interaction behavior sequences when x is the first service interaction index, sigma is the standard deviation of each information interaction behavior sequence corresponding to the first service interaction index.

Assuming that after the unified rule conversion processing is performed on the first service interaction index m1, the second service interaction index m2 and the service coverage information m3, a first unified rule value m1 'of the first service interaction index m1, a second unified rule value m 2' of the second service interaction index m2 and a third unified rule value m3 ', a, b and c corresponding to the service coverage information m3 are obtained, respectively representing influence factors of scores of the first service interaction index, the second service interaction index and the service coverage information (a, b and c are all greater than 0 and less than 1), a service interest value m of the information interaction behavior sequence is a m 1' + b m2 '-c m 3', so that the service interaction processes corresponding to the information interaction behavior sequence can be ranked according to the service interest value m, and a total rank arrangement of the service interaction processes corresponding to the interaction behavior sequence is obtained, wherein, the higher the score of m, the lower the order ranking, and the higher the possibility that the corresponding business interaction process is a business interest business interaction process. Therefore, the user portrait mining in the service interaction process can be carried out according to the service interest values of the information interaction behavior sequence corresponding to each service interaction process. For example, the method for mining the user portrait in the service interaction process may be: the business interaction process corresponding to the information interaction behavior sequences with higher business interest values and preset quantity comprises the step of mining a business interaction label as a mined user portrait, wherein the preset quantity can be a fixed value and can also be obtained by calculation according to the preset business interest ratio and the total quantity of all the information interaction behavior sequences; or calculating a difference value between the business interest value of each information interaction behavior sequence and the business interest value with the highest score, and when the difference value is smaller than a preset difference value threshold value, taking a business interaction process corresponding to the information interaction behavior sequence as mined user portrait and mining a business interaction label.

In addition, after the total sequence arrangement of the business interaction process is obtained, the business interaction labels including the information interaction behavior sequence with the highest sequence arrangement can be used as high-value user portraits, the business interaction labels including the information interaction behavior sequence with the lower sequence arrangement can be used as low-value user portraits, analysis can be performed on the business interaction process with the lower sequence arrangement, for example, the difference of the target micro-service starting numbers of all time periods in one day and the trend variation difference of a line graph in the high-value user portraits and the low-value user portraits are contrastively analyzed, and a great problem may exist if the target micro-service starting numbers of all time periods in one day in the low-value user portraits are basically balanced. Differences in the number of target microservice activations for each day of the week, etc. may also be analyzed.

In the embodiment of the disclosure, an information interaction behavior sequence corresponding to each service interaction process is obtained, wherein each information interaction behavior sequence comprises interaction behavior data of at least two target interaction information; determining a first service interaction index of each information interaction behavior sequence according to the information interaction behavior of each target interaction information when the target micro service is called for the first time; determining a second service interaction index of each information interaction behavior sequence according to the information interaction behavior of each target interaction information when the target micro service is called for the first time in each interaction stage; determining service coverage information of each information interaction behavior sequence according to a calling service node when each target interaction information calls the target micro service each time; and further, according to the first service interaction index, the second service interaction index and the service coverage information of the information interaction behavior sequence corresponding to each service interaction process, user portrait mining in the service interaction process is carried out. The method integrates a first service interaction index based on single information interaction behavior service interest detection, a second service interaction index based on multiple information interaction behavior service interest detection, and a service coverage information evaluation method based on service node calling and frequency rule reporting of information interaction behavior, wherein the first service interaction index and the second service interaction index analyze a service interaction process based on interest characteristics of the information interaction behavior, the service coverage information analyzes the service interaction process through service use characteristics of a user, and the interest characteristics and the service use characteristics of the information interaction behavior can be simultaneously integrated by combining the first service interaction index and the second service interaction index, so that the accuracy of user portrait mining is improved.

The method and the device for pushing the micro service in the service interaction process sequence order sort the value of the pushing service interaction process of the target micro service based on the information interaction behavior, and the final service interaction process sorting result integrates the sequence sorting results of three service interaction processes, namely a service interest detection result based on a single information interaction behavior, a service interest detection result based on a plurality of information interaction behaviors and service coverage information. The method and the device can output the sequence arrangement of the business interaction process, can output the weighted summary score of each business interaction process, and assist in further judging the value of the business interaction process. For the push service interaction process with low value identified, whether to reduce or stop the weight or frequency of the push service interaction process with low value can be determined after secondary analysis, and meanwhile, the weight or frequency of push update configuration can be properly increased for the push service interaction process with high value.

It is noted that while for simplicity of explanation, the foregoing method embodiments have been described as a series of acts or combination of acts, it will be appreciated by those skilled in the art that the present disclosure is not limited by the order of acts, as some steps may, in accordance with the present disclosure, occur in other orders and concurrently. Further, those skilled in the art should also appreciate that the embodiments described in the specification are preferred embodiments and that the acts and modules referred to are not necessarily required for the disclosure.

On the basis of the above description, some terms involved in the following embodiments of the present disclosure are explained below.

Resource pushing configuration information: and configuring item factors for influencing the pushing behavior of the subscription service item, such as pushing frequency setting, pushing hierarchy setting, pushing interval period setting, pushing duration range setting and the like. Service use trend characteristics: the method is used for representing the service use stage of the service process of the subscribed service item or representing the increase or slow-down situation of the service use of the subscribed service item, such as the service usage amount of the subscribed service item, the service usage rate of the subscribed service item, the service update amount of the subscribed service item, and the like. Dynamic service scenario features: the method is used for representing dynamic scene conditions except for big data information service, such as global pushing frequency, global scene utilization rate and the like. Static service scenario features: the method is used for representing static scene conditions in the big data information service, such as the passive push frequency of the big data information service, the push level in the big data information service and the like.

In the following embodiments, the dynamic service scene characteristic of the first online mode stage may be real-time dynamic scene information acquisition obtained by the third application, that is, the dynamic service scene characteristic of the first online mode stage is a real dynamic scene state of the first online mode stage, or may be previous synchronous dynamic scene data, that is, the dynamic service scene characteristic of the first online mode stage is a dynamic scene state simulated by the first online mode stage. The first on-line mode stage is a past on-line mode stage relative to the second on-line mode stage, and units of the first on-line mode stage and the second on-line mode stage can be flexibly set according to actual development and a service iteration cycle. The static service scene characteristics are obtained by sensing the pushing behavior in the big data information service through the service script, and the service use trend characteristics are obtained by sensing the trend of subscribing service items in the big data information service through the service script.

Before the foregoing step S110, the method provided in this embodiment may further include the following steps.

Step S101, obtaining service use trend characteristics of subscribed service items in the first online mode stage, static service scene characteristics of the big data information service in the first online mode stage, and dynamic service scene characteristics of the big data information service in the first online mode stage.

For example, the machine learning system 100 may obtain the static service scenario characteristic and the service usage trend characteristic of the past online mode stage (the first online mode stage) relative to the current online mode stage (the second online mode stage) through the service script in the big data information service, obtain the dynamic service scenario characteristic of the first online mode stage and the dynamic service scenario characteristic of the second online mode stage through the third application, automatically generate the configuration instruction for the big data information service based on the service usage trend characteristic of the first online mode stage, the static service scenario characteristic of the first online mode stage and the dynamic service scenario characteristic of the first online mode stage, and obtain the service usage trend characteristic of the first online mode stage, the static service scenario characteristic of the first online mode stage and the dynamic service scenario characteristic of the first online mode stage by analyzing the configuration instruction for the big data information service, and calling a target training model to perform dynamic optimization of resource pushing configuration information based on the service use trend characteristic of the first online mode stage, the static service scene characteristic of the first online mode stage and the dynamic service scene characteristic of the first online mode stage.

Step S102, a target training model is called based on the service use trend characteristic of the first online mode stage, the static service scene characteristic of the first online mode stage and the dynamic service scene characteristic of the first online mode stage to obtain resource pushing configuration information for configuring big data information service in the second online mode stage, wherein the second online mode stage is in the next development and updating stage of the first online mode stage.

For example, the service usage trend feature of the first online mode stage, the static service scenario feature of the first online mode stage, and the dynamic service scenario feature of the first online mode stage are input into the target training model, so that the target training model learns the following strategies: the resource pushing configuration information and the dynamic service scene characteristics of the big data information service are used as the static service scene characteristics of the big data information service, the static service scene characteristics of the big data information service are used as the service use trend characteristics of the subscribed service items, prediction processing is carried out through a target training model, the resource pushing configuration information used for configuring the big data information service in the second on-line mode stage, namely the resource pushing configuration information used for configuring the big data information service in the next on-line mode stage is obtained, the pushing behavior of the subscribed service items in the big data information service is dynamically adjusted and controlled in combination with the dynamic service scene characteristics, and intelligent and accurate prediction of the pushing behavior of the subscribed service items is further achieved.

Illustratively, in a separate embodiment, step S102 may be implemented by steps S1021-S1023: performing the following processing based on the target training model: in step S1021, based on the service usage trend characteristic of the first on-line mode stage, determining a preset service usage trend characteristic that satisfies the push update requirement in the second on-line mode stage; in step S1022, determining a static service scene characteristic of the second online mode stage when the preset service usage trend characteristic is realized; in step S1023, the resource pushing configuration information for configuring the big data information service in the second on-line mode stage is determined based on the linkage relationship between the dynamic service scenario feature of the second on-line mode stage and the static service scenario feature of the second on-line mode stage where the resource pushing configuration information for configuring the big data information service jointly acts on.

For example, the machine learning system 100 obtains the service usage trend feature of the first online mode stage, the static service scenario feature of the first online mode stage, and the dynamic service scenario feature of the first online mode stage by analyzing the configuration command for the big data information service, and calling a target training model to perform prediction processing, based on the strategy learned by the target training model, firstly performing prediction processing on service use trend characteristics based on the service use trend characteristics of the first on-line mode stage to obtain preset service use trend characteristics meeting the push updating requirement in the second on-line mode stage, the push update requirement may be set in advance according to an actual scene, for example, the push update requirement is that the service conversion rate of the subscription service item is maximized, or that the net worth of the service brought by the subscription service item is maximized. After the preset service use trend characteristic of the second on-line mode stage is obtained, the static service scene characteristic of the second on-line mode stage when the preset service use trend characteristic is realized is determined, finally, the dynamic service scene characteristic of the second on-line mode stage, which is learned based on the target training model, and the resource pushing configuration information used for configuring the big data information service in the second on-line mode stage jointly act on the linkage relation of the static service scene characteristic of the second on-line mode stage, and the resource pushing configuration information used for configuring the big data information service in the second on-line mode stage is determined, so that the pushing behavior of the subscribed service items in the big data information service is intelligently controlled.

In a single embodiment, the service usage trend characteristic of the first online mode stage is that the service usage rate of the subscribed service item is 50%, the service usage trend characteristic of the first online mode stage is predicted based on the pushing update requirement of the maximum net value of the service brought by the subscribed service item, the preset service usage trend characteristic of the second online mode stage is obtained as the service usage rate of 75%, in order to realize the preset service usage trend characteristic of the second online mode stage, the static service scenario characteristic of the second online mode stage is that the big data information service pushing frequency reaches 8 times/second, the pushing tier reaches 5 span levels, and the dynamic service scenario characteristic of the second online mode stage is obtained through the dynamic scenario information acquisition of the third party application as the global pushing frequency is 5 times/second, based on the linkage relationship that the dynamic service scene characteristics of the second on-line mode stage and the resource pushing configuration information for configuring the big data information service in the second on-line mode stage jointly act on the static service scene characteristics of the second on-line mode stage, the resource pushing configuration information of the second on-line mode stage is determined to be that the pushing frequency is set to be started to reach 8 times/second, the pushing level is set to be started to reach 5 span levels, and the pushing interval period is set to be 1 hour.

In a separate embodiment, the target training model includes a first processing unit, a second processing unit, and a feature processing unit; performing feature association processing on the service use trend feature of the first on-line mode stage based on the first processing unit to obtain a preset service use trend feature which meets the push updating requirement in the second on-line mode stage; correspondingly, based on the characteristic association relationship between the static service scene characteristic of the big data information service and the service use trend characteristic of the subscribed service item, the second processing unit performs characteristic association processing on the preset service use trend characteristic of the second on-line mode stage to obtain the static service scene characteristic of the second on-line mode stage when the preset service use trend characteristic of the second on-line mode stage is realized; correspondingly, based on the feature association relationship between the resource pushing configuration information and the dynamic service scene feature of the big data information service, which are included in the feature processing unit, and the static service scene feature of the big data information service, the static service scene feature of the second on-line mode stage is mapped to the resource pushing configuration information used for configuring the big data information service in the second on-line mode stage.

In an independent embodiment, the service usage trend feature of the first online mode stage can be input into a first processing unit in a target training model, the service usage trend feature of the first online mode stage is subjected to feature association processing through the first processing unit to obtain a preset service usage trend feature which meets the push updating requirement in the second online mode stage, the preset service usage trend feature is input into a second processing unit, the feature association relationship between the static service scene feature of the big data information service and the service usage trend feature of the subscribed service item is learned through the second processing unit, the feature association processing is carried out on the preset service usage trend feature of the second online mode stage to obtain the static service scene feature of the second online mode stage when the preset service usage trend feature of the second online mode stage is realized, and inputting the static service scene characteristics of the second on-line mode stage into the characteristic processing unit, and mapping the static service scene characteristics of the second on-line mode stage into the resource pushing configuration information for configuring the big data information service in the second on-line mode stage through the characteristic association relation between the resource pushing configuration information and the dynamic service scene characteristics of the big data information service learned by the characteristic processing unit and the static service scene characteristics of the big data information service. Therefore, the resource pushing configuration information of the second online mode stage can be predicted through the multilayer units in the target training model, intelligent and accurate prediction of the pushing behavior of the subscription business items is further achieved, and adjustment of the pushing behavior of the subscription business items in the big data information service in a manual regulation mode is avoided.

In a separate embodiment, the target training model includes a first processing unit, a second processing unit, and a feature processing unit; performing feature association processing on the service use trend feature of the first on-line mode stage based on the first processing unit to obtain a preset service use trend feature which meets the push updating requirement in the second on-line mode stage; correspondingly, based on the characteristic incidence relation between the static service scene characteristic of the big data information service and the service use trend characteristic change of the subscribed service item, the characteristic conversion processing is carried out on the characteristic difference between the service use trend characteristic of the first on-line mode stage and the preset service use trend characteristic of the second on-line mode stage, so as to obtain the static service scene characteristic of the second on-line mode stage when the preset service use trend characteristic of the second on-line mode stage is realized; correspondingly, based on the feature association relationship between the resource pushing configuration information and the dynamic service scene feature of the big data information service, which are included in the feature processing unit, and the static service scene feature of the big data information service, the static service scene feature of the second on-line mode stage is mapped to the resource pushing configuration information used for configuring the big data information service in the second on-line mode stage.

For example, the service usage trend feature of the first online mode stage may be input to a first processing unit in the target training model, the service usage trend feature of the first online mode stage is subjected to feature association processing by the first processing unit to obtain a preset service usage trend feature satisfying the push update requirement in the second online mode stage, the preset service usage trend feature is input to a second processing unit, a feature association relationship between the static service scenario feature of the big data information service learned by the second processing unit and the service usage trend feature change of the subscribed service item is obtained, the feature conversion processing is performed on the feature difference between the service usage trend feature of the first online mode stage and the preset service usage trend feature of the second online mode stage to obtain the static service scenario feature of the second online mode stage when the preset service usage trend feature of the second online mode stage is realized, the second processing unit inputs the static service scene characteristics of the second on-line mode stage into the characteristic processing unit, and the static service scene characteristics of the second on-line mode stage are mapped into the resource pushing configuration information for configuring the big data information service in the second on-line mode stage through the characteristic incidence relation between the resource pushing configuration information and the dynamic service scene characteristics of the big data information service, which are learned by the characteristic processing unit, and the static service scene characteristics of the big data information service. Therefore, the resource pushing configuration information of the second online mode stage can be predicted through the multilayer units in the target training model, intelligent and accurate prediction of the pushing behavior of the subscription business items is further achieved, and adjustment of the pushing behavior of the subscription business items in the big data information service in a manual regulation mode is avoided.

In an independent embodiment, the service usage trend characteristic of the first online mode stage is that the service usage rate of the subscribed service item is 50%, the preset service usage trend characteristic of the second online mode stage is that the service usage rate is 75%, the service usage trend characteristic of the subscribed service item changes to be 25% of the service usage rate, and in order to realize the change of the service usage trend characteristic of the subscribed service item, the static service scenario characteristic of the second online mode stage is that the pushing frequency of the big data information service reaches 8 times/second and the pushing hierarchy reaches 5 span levels, so that the relationship between the static service scenario characteristic of the big data information service and the change of the service usage trend characteristic of the subscribed service item in the actual growth process of the subscribed service item is learned through the second processing unit, and the relation between the static service scenario characteristic of the second online mode stage and the change of the service usage trend characteristic of the subscribed service scenario characteristic of the second online mode stage when the preset service usage trend characteristic of the second online mode stage is realized is accurately So as to accurately locate the resource pushing configuration information for configuring the big data information service in the second on-line mode stage.

In a separate embodiment, the feature processing unit includes a first feature extraction node, a second feature extraction node, a fully connected node, and a third feature extraction node; mapping the static service scenario features of the second on-line mode stage to resource push configuration information for configuring big data information service in the second on-line mode stage, comprising: performing feature extraction on the dynamic service scene features of the mode stage on the second line based on the first feature extraction node to obtain a first feature vector corresponding to the dynamic service scene features; performing feature extraction on the static service scene features of the mode stage on the second line based on a second feature extraction node to obtain a second feature vector corresponding to the static service scene features; determining a feature difference of the second feature vector and the first feature vector based on the fully connected nodes; and performing feature extraction on the feature difference based on the third feature extraction node to obtain resource pushing configuration information for configuring the big data information service in the second on-line mode stage.

For example, after the second processing unit inputs the static service scene features of the second on-line mode stage into the feature processing unit, based on the learned feature association relationship between the resource pushing configuration information and the dynamic service scene features of the big data information service and the static service scene features of the big data information service, feature extraction is performed on the dynamic service scene features of the second on-line mode stage through the first feature extraction node in the feature processing unit to obtain a first feature vector (i.e. a first convolution vector) corresponding to the dynamic service scene features, feature extraction is performed on the static service scene features of the second on-line mode stage through the second feature extraction node in the feature processing unit to obtain a second feature vector (i.e. a second convolution vector) corresponding to the static service scene features, and then the second feature vector is differentiated from the first feature vector through the full-connection node, and finally, performing feature extraction on the feature difference through a third feature extraction node to obtain resource pushing configuration information for configuring the big data information service in the second on-line mode stage.

Step S103, applying the resource pushing configuration information to the big data information service in the second on-line mode stage.

For example, the machine learning system 100 sends a configuration instruction for a big data information service to a server, the server receives the configuration instruction for the big data information service, calls a target training model to perform prediction processing, obtains resource pushing configuration information of a second on-line mode stage, and feeds the resource pushing configuration information of the second on-line mode stage back to the machine learning system 100, the machine learning system 100 applies the resource pushing configuration information to the big data information service in the second on-line mode stage through a service device, so as to convert a static service scenario feature of the first on-line mode stage into a static service scenario feature of the second on-line mode stage, and convert a service usage trend feature of the first on-line mode stage into a service usage trend feature of the second on-line mode stage under the combined action of the static service scenario feature of the second on-line mode stage and the dynamic service scenario feature of the second on-line mode stage, therefore, the growth of the subscription business items in the big data information service is automatically controlled.

In a separate embodiment, step S104-step S105 may also be included: in step S104, constructing a labeled feature object of the target training model based on the service usage trend feature of the subscribed service item in the big data information service at the multiple online mode stages, the static service scene feature of the big data information service at the multiple online mode stages, and the dynamic service scene feature of the big data information service at the multiple online mode stages; in step S105, a target training model is trained based on the labeled feature object, so as to obtain a target training model for predicting the resource pushing configuration information.

For example, the characteristics of the current online mode stage (including the traffic usage trend characteristic, the static service scenario characteristic, and the dynamic service scenario characteristic) are taken as the characteristics of the past online mode stage of the next online mode stage. Features (including a business use trend feature, a static service scene feature and a dynamic service scene feature) in a business iteration stage of a business service process for subscribing business items in the big data information service can be used as a labeled feature object, a training target model, namely a plurality of past online mode stages belong to one business iteration stage of the business service process for subscribing business items, for example, the business service process period for subscribing business items is 30 stages, and the resource pushing configuration information is updated by taking the stages as a unit, so that the number of the past online mode stages is 30, namely, the 1 st stage, the 2 nd stage and the … th 30 th stage.

For example, the plurality of past online mode phases may be 1 st, 2 nd and 3 rd online mode phases, and the 1 st online mode phase is any one of the 1 st service iteration phases for subscribing the service item, the 2 nd online mode phase is any one of the 2 nd service iteration phases for subscribing the service item, and the 3 rd online mode phase is any one of the 3 rd service iteration phases for subscribing the service item.

In a separate embodiment, constructing the labeled feature object of the target training model based on the business usage trend feature of the subscribed business items in the big data information service in the multiple online mode stages, the static service scene feature of the big data information service in the multiple online mode stages, and the dynamic service scene feature of the big data information service in the multiple online mode stages includes: performing the following for any of a plurality of on-line mode phases: acquiring service use trend characteristics of a next online mode stage of the online mode stage, static service scene characteristics of the next online mode stage of the online mode stage and dynamic service scene characteristics of the next online mode stage of the online mode stage; combining service use trend characteristics of a subscribed service item in a big data information service in a past online mode stage, static service scene characteristics of the big data information service in the past online mode stage and dynamic service scene characteristics of the past online mode stage into a first feature vector of the past online mode stage; combining the service use trend characteristic of the next online mode stage of the online mode stage, the static service scene characteristic of the next online mode stage of the online mode stage and the dynamic service scene characteristic of the next online mode stage of the online mode stage into a second feature vector of the next online mode stage; constructing a labeled feature object of a past online mode stage based on a first feature vector of the past online mode stage, resource pushing configuration information for configuring big data information service in the past online mode stage and a second feature vector of a next past online mode stage; and fusing the labeled characteristic objects of the plurality of passing online mode stages to obtain the labeled characteristic object of the target training model.

For example, when the traffic usage trend feature of a certain past online mode phase is x1, the static service scenario feature is y1, and the dynamic service scenario feature is z1, the feature vector of the past online mode phase is S1[ x1, y1, z1 ]. When the traffic usage trend feature of the next online mode phase of the online mode phase is x2, the static service scene feature is y2, and the dynamic service scene feature is z2, the feature vector of the next online mode phase of the online mode phase is S2[ x2, y2, z2 ]. When the resource pushing configuration information for configuring the big data information service in the past online mode stage is a1, the service usage trend feature S1 of the past online mode stage, the resource pushing configuration information for configuring the big data information service in the past online mode stage is a1, and the feature vector S2 of the next past online mode stage in the past online mode stage are combined as the labeled feature object of the past online mode stage [ S1, a1, S2 ]. When there are N past online mode stages, the feature objects are marked as [ S1, a1, S2], …, [ sN-1, aN-1, sN ], [ sN, aN, termination state ], wherein N is a natural number greater than 2, and the termination state represents a training termination condition.

In the process of dynamically adjusting and controlling the pushing behavior of the subscription business items in the big data information service, the marked feature objects in the past online mode stage can be stored in the extended subscription area to collect the marked feature objects, so that when the number of the marked feature objects in the past online mode stage in the extended subscription area reaches a set threshold value, a plurality of marked feature objects in the past online mode stage are obtained from the extended subscription area, and the target training model is trained based on the marked feature objects in the past online mode stage.

In a separate embodiment, training a target training model based on a labeled feature object of a business service process of a subscription business item in a big data information service to obtain the target training model for predicting resource pushing configuration information includes: constructing a model updating reference function of the target training model based on the labeled characteristic object of any one past online mode stage in the labeled characteristic objects and the labeled evaluation information of the past online mode stage; and updating parameters of the target training model until the model updating reference function is converged, and taking the updated parameters of the target training model when the model updating reference function is converged as the parameters of the target training model for predicting the resource pushing configuration information.

In a separate embodiment, before constructing the model update reference function of the target training model, the method further includes: calling a target training model to perform prediction processing based on a labeled feature object in any past online mode stage in the labeled feature objects to obtain prediction resource pushing configuration information for configuring big data information service in the past online mode stage; pushing configuration information based on the predicted resources to obtain predicted evaluation information of a past online mode stage; and constructing a model updating reference function of the target training model based on the labeled characteristic object of the mode stage on the previous line, the prediction evaluation information of the mode stage on the previous line and the labeled evaluation information of the mode stage on the previous line.

For example, in the training stage, a target training model is called to perform prediction processing, prediction resource pushing configuration information used for configuring a big data information service in a past online mode stage is obtained, and prediction evaluation information of the past online mode stage is obtained based on the prediction resource pushing configuration information, for example, based on the prediction resource pushing configuration information, service use trend characteristics which can be reached by a subscription service item in the big data information service are predicted, and service use trend prediction information brought by the service use trend characteristics which can be reached by the subscription service item is used as the prediction evaluation information; predicting service use trend characteristics which can be achieved by a subscription service item in the big data information service based on the predicted resource pushing configuration information, taking service use trend predicted information brought by the service use trend characteristics which can be achieved by the subscription service item as predicted evaluation information, predicting quoted service information associated with indexes required by the resource pushing configuration information, and taking the characteristic difference between the service use trend predicted information and the quoted service information as marked evaluation information. And finally, constructing a model updating reference function of the target training model based on the labeled characteristic object of the mode stage on the previous line, the prediction evaluation information of the mode stage on the previous line and the labeled evaluation information of the mode stage on the previous line.

In an independent embodiment, a first processing unit in a target training model is used for performing feature association processing on service use trend characteristics of a past online mode stage to obtain preset service use trend characteristics meeting push updating requirements in a next past online mode stage, and static service scene characteristics of the next past online mode stage are mapped into predicted resource push configuration information for configuring large data information services in the next past online mode stage based on feature association relations between resource push configuration information and dynamic service scene characteristics of large data information services and static service scene characteristics of the large data information services, wherein the resource push configuration information and the dynamic service scene characteristics are included in the feature processing unit in the target training model.

For example, after determining the value of the model update reference function (e.g., but not limited to, cross entropy loss function, etc.) of the target training model based on the tagged feature object of any past online mode stage, the predicted evaluation information of the past online mode stage, and the tagged evaluation information of the past online mode stage (e.g., net business value actually brought by the subscribed business item in the past online mode stage, or business transformation rate actually brought by the subscribed business item in the past online mode stage), it may be determined whether the value of the model update reference function of the target training model exceeds a preset threshold, when the value of the model update reference function of the target training model exceeds the preset threshold, an error signal of the target training model is determined based on the model update reference function of the target training model, and the error information is propagated back in the target training model, and updating the model parameters of each layer in the process of propagation.

In a separate embodiment, before constructing the model update reference function of the target training model, the method further includes: the method comprises the steps of obtaining service use trend characteristics of a next online mode stage in a labeled characteristic object of the online mode stage, calling a subscription service project simulator model based on the service use trend characteristics of the next online mode stage to determine service use trend estimation information brought by a service process of a subscription service project in the online mode stage, and taking the service use trend estimation information as labeled evaluation information of the online mode stage. For example, the actual service conversion rate of the subscribed service item in the past online mode stage is used as the labeled evaluation information of the past online mode stage.

In a separate embodiment, before constructing the model update reference function of the target training model, the method further includes: acquiring service use trend characteristics of a next online mode stage in a marking characteristic object of the online mode stage, and calling a subscription service project simulator model based on the service use trend characteristics of the next online mode stage to determine service use trend estimation information brought by a service process of a subscription service project in the online mode stage; acquiring resource pushing configuration information for configuring big data information service in a past online mode stage; calling a subscription business project simulator model based on resource pushing configuration information for configuring big data information service in a past online mode stage to determine reference business information associated with indexes required by the resource pushing configuration information, and taking business use trend prediction information as marking evaluation information of the past online mode stage, wherein the method comprises the following steps: and taking the characteristic difference between the service use trend estimated information and the quoted service information as the marking evaluation information of the past online mode stage. For example, the net business value actually brought by the subscription business item in the past online mode stage is used as the labeled evaluation information of the past online mode stage.

In a possible embodiment, the foregoing step S103 can be further realized by the following exemplary steps.

Step S1031, obtaining the update instruction of the resource pushing configuration information for the big data information service.

In this embodiment, the machine learning system 100 may receive an update instruction sent by the front-end service node, where the update instruction is used to instruct to perform a push status update on the current push information text information in the resource push configuration information.

Step S1032, when the current push information text message is in the first push configuration state, determining an update mode corresponding to the current push information text message and all update service microservices in the update mode that are in the update activation state, and requesting all update service microservices to respond to the update instruction, and executing push state update for the current push information text message.

Step S1033, when the current push information text message is in the second push configuration state, determining a current push configuration instruction of the current push information text message in the resource push configuration message, and when the current push configuration instruction meets a preset update level, executing a push state update for the current push information text message.

In this embodiment, a status information identifier is stored in the current push information text message, and the status information identifier is used to indicate the status of the current push information text message; for example, when the status information identifier indicates that the current push information text message is in the first push configuration state, the deep learning system 100 determines that the current push information text message is in the complete push activation state, and at this time, the deep learning system 100 may determine an update mode corresponding to the current push information text message and all update services in the update mode that are in the update activation state.

Wherein, an update tag can be stored in the current push information text message, and the update tag can be used for indicating a corresponding update mode; the deep learning system 100 may store an update label mapping table, where the update label mapping table records a correspondence between a plurality of update labels and a plurality of update modes; the deep learning system 100 may obtain an update mode corresponding to the update tag information stored in the current push information text information by querying the update tag mapping table; and, the deep learning system 100 may send query information to the application service of the determined update mode to request response information returned by all the update service microservices in the update mode.

Next, the deep learning system 100 may determine the working state of each update service microservice according to the condition that each update service microservice in the update mode returns response information; for example, the deep learning system 100 may determine that the update service microservice that feeds back the response information within the set time is in the update activated state, and the update service microservice that does not feed back the response information beyond the set time is not in the update activated state; thus, after the deep learning system 100 determines all the update service microservices in the update activation state in the update mode, the deep learning system 100 may request all the update service microservices to respond to the update request, and perform the update of the push state with respect to the current push information text information.

In addition, when the status information identifier indicates that the current pushed information text information is in the second pushing configuration state, the deep learning system 100 determines that the current pushed information text information is in a partial pushing activation state, that is, the current pushed information text information contains some data which is not pushed and activated, at this time, the deep learning system 100 can determine a current pushing configuration instruction of the current pushed information text information in the resource pushing configuration information, and look up a preconfigured update level policy table, where update levels corresponding to multiple signaling are recorded in the update level policy table; when the deep learning system 100 determines the update level corresponding to the current push configuration instruction by searching the update level policy table, and when the deep learning system 100 determines that the update level corresponding to the current push configuration instruction meets the preset update level, the deep learning system 100 determines that the current push information text information meets the update requirement, and then the deep learning system 100 executes push state update for the current push information text information.

It can be seen that, in the solution provided by the present disclosure, by obtaining an update instruction of resource push configuration information for a big data information service, where the update instruction is used to instruct to perform push status update on current push information text information in the resource push configuration information; under the condition that the current pushed information text information is in a first pushing configuration state, determining an updating mode corresponding to the current pushed information text information and all updating service micro-services in an updating activation state in the updating mode, requesting all updating service micro-services to respond to an updating instruction, and executing pushing state updating aiming at the current pushed information text information; in addition, under the condition that the current push information text message is in the second push configuration state, a current push configuration instruction of the current push information text message in the resource push configuration message is determined, and when the current push configuration instruction meets a preset update level, the push state update is executed aiming at the current push information text message. Therefore, the push reliability of the push information text information is improved and data pollution is avoided by judging two conditions that the current push information text information is in the first push configuration state or the second push configuration state and responding to the update instruction of the current push information text information according to different conditions to execute push state update.

In a separate embodiment, for different working states of the update mode, after the machine learning system 100 executes step S1031, it may also obtain the resource pushing description information of the resource pushing configuration information.

Next, the machine learning system 100 may obtain the operating status of the update mode in the manner described above; moreover, when the instruction indicating that the update mode is in the update maintenance mode is obtained, the machine learning system 100 may switch the current push information text message from the first push configuration state to the second push configuration state, and obtain the current push configuration instruction of the current push information text message in the resource push configuration information according to the resource push description information, and then execute step S1033.

In this embodiment, the current push configuration instruction may be recorded in a specified field of the resource push description information.

In addition, when the machine learning system 100 obtains the information indicating that the update mode is in the update non-maintenance mode, the current push information text information is switched from the second push configuration state to the first push configuration state, and the management content description data in the resource push configuration information is obtained according to the resource push description information; thereby performing step S1032.

In a separate embodiment, the machine learning system 100 may adopt the following scheme when executing step S1033 to determine the current push configuration instruction of the current push information text message in the resource push configuration information:

first, the machine learning system 100 may use an information text configuration source of the current pushed information text information in the resource pushing configuration information as a target configuration source, and determine a traversal configuration source range by using a preset information text configuration source range in the current pushed information text information as a fixed range. In this embodiment, the preset information text configuration source range may be recorded in the preset configuration source range in the current push information text message. Next, the machine learning system 100 may determine a high heat range of the configuration source of the current push information text message from the traversal configuration source range according to the heat range of the configuration source of the current push information text message; wherein, the high heat range of the allocation source is the high heat range of the push allocation command stored in the current push information text message. In this embodiment, the configuration source heat range may be stored in the metadata of the current push info text message. The machine learning system 100 may then search for the current push configuration instructions for the current push informational text message from the configuration source high heat range. Next, when the current push configuration instruction is not searched from the high-heat range of the configuration source, the machine learning system 100 may use the information text configuration source of the current push information text information in the resource push configuration information as the target configuration source, use the corresponding information text index of the current push information text information as the target operation object, and obtain a plurality of push information text indexes by dividing according to the set index categories. Then, the machine learning system 100 may obtain an index type code corresponding to each of the plurality of pushed information text indexes, and merge all the index type codes according to the order of the pushed information text indexes corresponding to each of the index type codes to generate a current pushing configuration instruction of the current pushed information text information.

In addition, in a separate embodiment, in order to evaluate the overall value of the big data information service, the information processing method may further include the following steps for the current pushed information text information corresponding to the big data information service:

first, the machine learning system 100 may obtain respective historical pushing behavior information of the current pushing information text information in a preset pushing cycle period and pushing behavior feedback information corresponding to each piece of historical pushing behavior information. In this embodiment, the predetermined push cycle period may be five cycle periods in the past for the big data information service; the historical push behavior information may include information referencing control information during the push behavior within each preset push cycle period. Next, the machine learning system 100 may perform push index change calculation on the current push information text information according to the historical push behavior information and the push behavior feedback information, so as to obtain push index change evaluation information corresponding to the current push information text information in each push cycle period. In this embodiment, the machine learning system 100 may perform service index increase calculation on the current pushed information text information according to the historical pushed behavior information and the pushed behavior feedback information by using a least square fitting, for example, so as to obtain the pushed index change evaluation information corresponding to the current pushed information text information in each pushing cycle period. Then, the machine learning system 100 may perform importance parameter calculation on each piece of push index change evaluation information to obtain at least one set of push index importance parameters. Next, the machine learning system 100 may determine current push index change evaluation information to be estimated and related push index change evaluation information of a period adjacent to the current push index change evaluation information; the relevant push index change evaluation information is push index change evaluation information of a previous period adjacent to the current push index change evaluation information. Then, the machine learning system 100 may determine, according to the push index importance parameter, an expected push effective conversion parameter corresponding to each of the current push index change evaluation information and the related push index change evaluation information. Next, the machine learning system 100 may estimate a push value of the current push information text message according to all the expected push effective transformation parameters and the push index change evaluation information corresponding to the expected push effective transformation parameters, so as to obtain expected push value data corresponding to the current push information text message. Then, the machine learning system 100 may obtain historical push value data corresponding to the historical push information text information, and determine current push expected data of the current push information text information according to the historical push value data; the historical push information text message is a previous push information text message corresponding to the current push information text message. Next, the machine learning system 100 may perform a push value degree calculation on the current push expected data according to the expected push value data to obtain a push value influence parameter of the current push information text information. Then, the machine learning system 100 may perform value estimation on the current pushed information text message according to the pushed value influence parameter to obtain a reference pushed value corresponding to the current pushed information text message.

Therefore, based on the scheme provided by the embodiment of the disclosure, the push value of the current push information text information is estimated, and the overall value of the current push information text information of the big data information service can be estimated based on the estimated push value, so that the index of the big data information service is objectively estimated, and the reliability of service expansion is improved.

As a possible implementation manner, in order to improve reliability of obtaining the historical pushed value data and prevent data from being leaked, the machine learning system 100 may adopt the following manner when obtaining the historical pushed value data corresponding to the historical pushed information text information:

first, the machine learning system 100 may determine a target historical information update item corresponding to the current pushed information text information, and obtain a historical information update data packet stored by the target historical information update item; wherein, the historical information updating data packet is a data packet of all historical push information text information stored in the target historical information updating item. Next, the machine learning system 100 may obtain the update sequence number of the history information update packet fed back by the target history information update item. The machine learning system 100 may then decode the historical information update packet using the update sequence number. Next, the machine learning system 100 may obtain a statistical period start node and a statistical period end node of the historical pushed information text information. Then, the machine learning system 100 may determine the target statistical period according to the statistical period start node and the statistical period end node. Next, the machine learning system 100 may obtain the target historical push subscription information of the decoded historical information update data packet in the target statistical period. Then, the machine learning system 100 can analyze the historical push value data corresponding to the current push information text information from the target historical push subscription information.

Fig. 3 is a schematic diagram of functional modules of a user portrait mining apparatus 300 based on a big data line mode according to an embodiment of the present disclosure, and the functions of the functional modules of the user portrait mining apparatus 300 based on a big data line mode are described in detail below.

The obtaining module 310 is configured to obtain an information interaction behavior sequence corresponding to each service interaction process in a big data information service of a service user of a service device, where the information interaction behavior sequence includes interaction behavior data of at least two target interaction information, and the interaction behavior data includes information interaction behavior of the target interaction information when the target micro service is called each time and a calling service node.

The first determining module 320 is configured to determine a service interaction indicator of each information interaction behavior sequence according to the information interaction behavior when each target interaction information invokes the target micro service, where the service interaction indicator is used to represent a service interest degree of the target interaction information in the information interaction behavior sequence when the target micro service is invoked.

The second determining module 330 is configured to determine the service coverage information of each information interaction behavior sequence according to the calling service node when each target interaction information calls the target microservice each time. The service coverage information is used for representing the service coverage degree of the target micro service called by the target interaction information in the information interaction behavior sequence.

And the mining module 340 is configured to mine the user portrait in the service interaction process according to the service interaction indicators and the service coverage information of the information interaction behavior sequence corresponding to each service interaction process.

Fig. 4 is a schematic diagram illustrating a hardware structure of a machine learning system 100 for implementing the above-described user portrait mining method based on patterns on a big data line according to an embodiment of the present disclosure, where as shown in fig. 4, the machine learning system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a transceiver 140.

In a specific implementation process, at least one processor 110 executes computer-executable instructions stored in the machine-readable storage medium 120, so that the processor 110 may execute the user portrait mining method based on the pattern on the big data line according to the above method embodiment, the processor 110, the machine-readable storage medium 120, and the transceiver 140 are connected through the bus 130, and the processor 110 may be configured to control the transceiving action of the transceiver 140, so as to perform data transceiving with the business service device 200.

For a specific implementation process of the processor 110, reference may be made to the above-mentioned method embodiments executed by the machine learning system 100, which implement the principle and the technical effect similarly, and the detailed description of the embodiment is omitted here.

In addition, the embodiment of the disclosure also provides a readable storage medium, wherein a computer execution instruction is preset in the readable storage medium, and when a processor executes the computer execution instruction, the user portrait mining method based on the mode on the big data line is realized.

Finally, it should be understood that the examples in this specification are only intended to illustrate the principles of the examples in this specification. Other variations are also possible within the scope of this description. Thus, in a separate embodiment, not limiting, alternative configurations of the embodiments of the present specification are contemplated as being consistent with the teachings of the present specification. Accordingly, the embodiments of the present description are not limited to only those embodiments explicitly described and depicted herein.

Claims

1. A user portrait mining method based on a big data online mode is applied to a machine learning system, the machine learning system is in communication connection with a plurality of business service devices, and the method comprises the following steps:

2. The method as claimed in claim 1, wherein the step of determining the service interaction indicator of each information interaction behavior sequence according to the information interaction behavior when the target micro service is called based on each target interaction information comprises:

determining a first service interaction index of each information interaction behavior sequence according to the information interaction behavior of each target interaction information when the target micro service is called for the first time; the first business interaction index is used for representing the business interest degree of target interaction information in the information interaction behavior sequence when the target micro service is configured;

determining a second service interaction index of each information interaction behavior sequence according to the information interaction behavior of each target interaction information when the target micro service is called for the first time in each interaction stage; the second business interaction index is used for representing the business interest degree of the target interaction information in the information interaction behavior sequence when the target micro service is started;

and determining the service interaction indexes of the information interaction behavior sequences according to the first service interaction indexes and the second service interaction indexes of each information interaction behavior sequence.

3. The method as claimed in claim 1, wherein the step of obtaining the information interaction behavior sequence corresponding to each service interaction process comprises:

acquiring interaction behavior data of each target interaction information, wherein the interaction behavior data comprise a service interaction process ID, and an information interaction behavior and a calling service node when the target interaction information calls the target micro service each time, and the service interaction process ID is used for indicating the target interaction information to obtain a service interaction process of the target micro service;

classifying the interaction behavior data of each target interaction information according to the service interaction process ID to obtain candidate information interaction behavior sequences corresponding to the service interaction processes indicated by each service interaction process ID respectively, wherein each candidate information interaction behavior sequence comprises interaction behavior data of at least two target interaction information, and all target interaction information in the same candidate information interaction behavior sequence have the same service interaction process ID;

and counting the data quantity of the interaction behavior data in each candidate information interaction behavior sequence, and taking the candidate information interaction behavior sequence with the data quantity more than or equal to a preset threshold value as the information interaction behavior sequence.

4. The method as claimed in claim 2, wherein the step of determining the first service interaction indicator of each information interaction behavior sequence according to the information interaction behavior when each target interaction information first calls the target microservice comprises:

taking the information interaction behavior of each target interaction information when the target micro service is called for the first time as the first information interaction behavior of the target interaction information;

performing feature processing on the first information interaction behavior of each target interaction information to obtain first interaction behavior features corresponding to each target interaction information;

inputting the first interaction behavior characteristics of each target interaction information into a service interest prediction network, and performing service interest characteristic analysis on the first interaction behavior characteristics of each target interaction information based on the service interest prediction network to obtain a first service interest value of each target interaction information;

and determining a first service interaction index corresponding to each information interaction behavior sequence based on the total amount of the target interaction information in each information interaction behavior sequence and a first service interest value of the target interaction information of which the first service interest value is greater than a first threshold value.

5. The method of claim 2, wherein determining the second service interaction indicator for each information interaction behavior sequence according to the information interaction behavior of each target interaction information when the target microservice is called for the first time in each interaction stage comprises:

taking the information interaction behavior corresponding to the first calling of the target micro service in each interaction stage of each target interaction information as a second information interaction behavior;

determining the target interaction information with the number of the second information interaction behaviors larger than or equal to a preset number threshold as candidate interaction information;

performing feature processing on all second information interaction behaviors of each candidate interaction information to obtain second interaction behavior features corresponding to each candidate interaction information;

inputting the second interaction behavior characteristics of each candidate interaction information into a service interest prediction network, and performing service interest analysis on the second interaction behavior characteristics of each candidate interaction information based on the service interest prediction network to obtain a second service interest value of each candidate interaction information;

and determining a second service interaction index corresponding to each information interaction behavior sequence based on the total number of the candidate interaction information in each information interaction behavior sequence and a second service interest value of the candidate interaction information of which the second service interest value is greater than a second threshold value.

6. The method as claimed in claim 1, wherein the step of determining the service coverage information of each information interaction behavior sequence according to the calling service node of each target interaction information calling the target microservice each time comprises:

according to each calling service node when each target interaction information calls the target micro service, counting the number of target interaction information for configuring the target micro service at a first calling service node and starting the target micro service at a second calling service node, the number of target interaction information for configuring the target micro service at the first calling service node, the number of target interaction information for starting the target micro service at the first calling service node and the second calling service node and the number of target interaction information for starting the target micro service at the first calling service node in each information interaction behavior sequence; the second calling service node is a calling service node behind the first calling service node;

calculating first service coverage information corresponding to each information interaction behavior sequence according to the quantity of target interaction information for configuring a target micro service at a first calling service node and starting the target micro service at a second calling service node and the quantity of target interaction information for configuring the target micro service at the first calling service node in each information interaction behavior sequence;

calculating second service coverage information corresponding to each information interaction behavior sequence according to the number of target interaction information of the target micro service started by the first calling service node and the second calling service node in each information interaction behavior sequence and the number of target interaction information of the target micro service started by the first calling service node;

and determining the service coverage information corresponding to each information interaction behavior sequence according to the first service coverage information and the second service coverage information corresponding to the information interaction behavior sequence.

7. The method as claimed in claim 2, wherein the step of mining the user portrait in the service interaction process according to the service interaction indicators and the service coverage information of the information interaction behavior sequence corresponding to each service interaction process comprises:

performing unified rule conversion processing on the first service interaction index, the second service interaction index and the service coverage information of each information interaction behavior sequence to obtain a first unified rule value corresponding to the first service interaction index, a second unified rule value corresponding to the second service interaction index and a third unified rule value corresponding to the service coverage information;

acquiring a preset first influence factor corresponding to the first service interaction index, a preset second influence factor corresponding to the second service interaction index and a preset third influence factor corresponding to the service coverage information;

according to the first influence factor, the second influence factor and the third influence factor, carrying out weighted calculation on a first uniform rule value, a second uniform rule value and a third uniform rule value of each information interaction behavior sequence to obtain a business interest value of each information interaction behavior sequence;

and mining the user portrait in the service interaction process according to the service interest value of the information interaction behavior sequence corresponding to each service interaction process.

8. The method for mining user portrait based on large data online mode according to any one of claims 1 to 7, wherein the method further comprises:

acquiring service use trend characteristics of a service item subscribed by a service user of the service equipment in a first on-line mode stage in a big data information service, static service scene characteristics of the big data information service in the first on-line mode stage and dynamic service scene characteristics of the big data information service in the first on-line mode stage;

calling a target training model based on the service use trend characteristic of the first online mode stage, the static service scene characteristic of the first online mode stage and the dynamic service scene characteristic of the first online mode stage to obtain resource pushing configuration information for configuring the big data information service in a second online mode stage, wherein the second online mode stage is in the next development and update stage of the first online mode stage;

and applying the resource pushing configuration information to the big data information service in the second on-line mode stage.

9. The method of claim 1, wherein the step of applying the resource push configuration information to the big data information service in the second online mode stage comprises:

in the second online mode stage, acquiring an update instruction of resource pushing configuration information aiming at the big data information service; wherein the update instruction is used for indicating that push state update is executed in the big data information service aiming at the current push information text information in the resource push configuration information;

when the current push information text information is in a first push configuration state, determining an update mode corresponding to the current push information text information and all update service micro-services in an update activation state in the update mode, requesting all the update service micro-services to respond to the update instruction, and executing push state update aiming at the current push information text information;

when the current push information text message is in a second push configuration state, determining a current push configuration instruction of the current push information text message in resource push configuration information, and when the current push configuration instruction meets a preset update level, executing push state update aiming at the current push information text message.

10. A machine learning system, the machine learning system comprising:

a memory for storing executable instructions;

a processor, configured to execute the executable instructions stored in the memory, to implement the method of user portrait mining based on patterns on big data lines of any of claims 1-9.