Detailed Description
The application provides a short message intelligent pushing method and system based on user analysis, which are used for solving the technical problems of low intelligent degree and poor pushing effect of short message pushing in the prior art.
Having introduced the basic principles of the present application, the technical solutions herein will now be clearly and fully described with reference to the accompanying drawings, it being apparent that the described embodiments are only some of the embodiments of the present application, and not all of the embodiments of the present application, and it is to be understood that the present application is not limited by the example embodiments described herein. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments of the present application without making any inventive effort, are intended to be within the scope of the present application. It should be further noted that, for convenience of description, only some, but not all of the drawings related to the present application are shown.
Example 1
As shown in fig. 1, the present application provides a method for intelligently pushing a short message based on user analysis, where the method includes:
s100: acquiring a target pushing information set, and analyzing the target pushing information set to acquire M target pushing subsets, wherein M is an integer greater than or equal to 1;
specifically, the target push information set refers to a set formed by short messages to be pushed, the information to be pushed is divided according to information types, M sets of different types are obtained, and the information types are determined according to the purpose of information pushing, including link pushing, daily information pushing, new business and the like. Classifying the short messages in the target pushing information set under M sets according to different types to obtain M target pushing sub-sets, wherein M is an integer greater than or equal to 1.
Further, the method comprises the following steps:
s110: extracting push information in the past time from a push database to obtain a historical push information set;
s120: the push information type is used as an index to extract information of the historical push information set, and an information type set is obtained;
s130: constructing an analysis decision tree according to the information type set and the historical push information set;
S140: and inputting the target pushing information set into the analysis decision tree to obtain M target pushing subsets.
In particular, the database is a warehouse that stores data. The storage space is large, and millions, tens of millions and hundreds of millions of data can be stored. The push database refers to a database obtained by summarizing and recording information pushed in a past time period, and the database comprises past push information, information to be pushed, objects to be pushed and the like. Taking time as a node to acquire push information in the past time, taking the type as an index tag, extracting data to obtain the pushed information conforming to the tag, and forming a history push information set; and carrying out information matching on the history push information set by taking push information types, namely links, daily information and the like as indexes, wherein the obtained information type set comprises daily information, weather push, new service recommendation and the like. Constructing an analysis decision tree according to the information type set and the historical push information set; inputting the target pushing information set into the analysis decision tree to obtain M target pushing subsets, and providing support for the construction of subsequent decision trees and the calculation of information entropy.
Further, as shown in fig. 2, the steps of the present application include:
s131: constructing a root node of the analysis decision tree according to the historical push information set;
s132: traversing the information type set to construct a plurality of leaf nodes of the analysis decision tree, and marking the leaf nodes in a type mode according to the information type;
s133: the historical push information set is input into the analysis decision tree to obtain an analysis set;
s134: and matching the information type set with the analysis set, and obtaining the constructed analysis decision tree if the matching success rate exceeds a preset threshold.
Specifically, the decision tree is a classical machine learning algorithm, a classification algorithm based on a tree structure for training a model from a given training dataset, and the classification criteria include entropy, coefficient of kunits, etc. Generally, a decision tree comprises a root node, internal nodes and leaf nodes. The analysis decision tree is a functional model for classifying the history push information, a root node of the analysis decision tree is constructed according to the history push information set, and the history push information is divided into: after accessing the data in the multiple information type sets of the analysis decision tree once, the data are divided according to different information types, wherein each type corresponds to one leaf node, namely, the multiple leaf nodes of the analysis decision tree are constructed according to the information type sets, and the corresponding leaf nodes are marked according to the information types, so that the information type corresponding to each leaf node can be identified, and the searching and distinguishing of the types corresponding to the leaf nodes are facilitated. The method comprises the steps that a historical push information set is input into a decision tree, the information types in the historical push information set are divided through the decision tree, and the division result is set to be an analysis set; and matching the information type set with the analysis set, and if the matching success rate exceeds a preset threshold, setting the threshold by a worker at the position by the self, thereby obtaining the constructed analysis decision tree. And the efficiency is improved for the information push.
Firstly, the number, the content and the like of the short messages to be pushed are obtained through definitely determining the short messages to be pushed; constructing a push database comprising short messages to be pushed and short messages of which the pushing is completed; extracting the short messages pushed in the past time period through a pushing database, namely collecting the short messages after pushing, and obtaining a history pushing information set; the method comprises the steps of taking push information types, including links, daily information push, new business and the like as indexes, extracting data from a history push information set to obtain data conforming to labels, and obtaining an information type set; then constructing a root node of an analysis decision tree according to the history push information set, inquiring all data in the information type set to construct a plurality of leaf nodes, namely, each type corresponds to one leaf node, marking the corresponding leaf node according to the information type, identifying the type corresponding to the leaf node more quickly, inputting the analysis set obtained in the decision tree through the history push information, matching the analysis set with the information type set, and obtaining the constructed analysis decision tree if the matching success rate exceeds a preset threshold; and finally, inputting the target pushing information set into a decision tree to obtain M target pushing subsets, namely, obtaining M sets of pushing information of different types, wherein M is an integer greater than or equal to 1. By dividing the types of the target push messages, the intelligent push messages are padded.
S200: information entropy calculation is carried out on the M target pushing subsets, and serialization processing is carried out on the M target pushing subsets according to calculation results, so that M target pushing subsequences are obtained;
specifically, the information entropy calculation means that the information quantity contained in each piece of information is calculated, and the information entropy calculation is carried out on M target pushing subsets to obtain a result, namely the pushing short message size; sequencing the M target pushing sub-sets, namely sequencing according to the information content, to obtain M target pushing sub-sequences; by sorting the information, a high-quality user can obtain high-quality information pushing, and the pushing effect of pushing the information is improved.
Further, the method comprises the following steps:
s210: traversing the M target pushing subsets to perform information theory coding operation to obtain M characteristic information entropy sets;
s220: randomly selecting a target pushing sub-set from the M target pushing sub-sets as a first target pushing sub-set, extracting a first characteristic information entropy set corresponding to the first target pushing sub-set, and training an input data size comparison model to obtain a first target pushing sub-sequence;
S230: randomly selecting a target pushing sub-set from the M target pushing sub-sets as a second target pushing sub-set, extracting a second characteristic information entropy set corresponding to the second target pushing sub-set, and training an input data size comparison model to obtain a second target pushing sub-sequence;
s240: randomly selecting a target pushing sub-set from the M target pushing sub-sets, taking the target pushing sub-set as an M-th target pushing sub-set, extracting an M-th characteristic information entropy set corresponding to the M-th target pushing sub-set, and training an input data size comparison model to obtain an M-th target pushing sub-sequence.
In particular, traversal of a tree is an important operation of a tree. The traversal refers to the access of information to all nodes in the tree, namely each node in the tree is accessed once and only once in turn, and the information entropy calculation in the information theory coding is performed by traversing the M target pushing subsets, so that the information entropy calculation formula in the information theory coding is as follows:
where t represents a random variable, corresponding to which is a set of all possible outputs, defined as a set of symbols, the output of the random variable is represented by t, p (i-t) represents an output probability function, and the greater the uncertainty of the variable, the greater the entropy.
In one possible embodiment, the information entropy sizes are ordered by establishing a data size comparison model, and after the characteristic information entropy set is input into the data size comparison model, the push subsequence is output. The data size comparison model is a functional model for comparing the sizes of the characteristic information entropy. The data size comparison model is processed by not replacing any selected characteristic information entropy from an obtained characteristic information entropy set to serve as a first characteristic information entropy, then not replacing any selected characteristic information entropy from the characteristic information entropy set to serve as a second characteristic information entropy, comparing the sizes of the first characteristic information entropy and the second characteristic entropy, if the first characteristic information entropy is larger than the second characteristic information entropy, arranging the first characteristic information entropy in the front of the second characteristic information entropy to obtain a first sorting result, then not replacing selected R characteristic information entropy from the characteristic information entropy set, comparing the one-by-one size with the characteristic information entropy in the R-1 sorting result, and merging the R sorting result into the R-1 sorting result to obtain an R sorting result, and obtaining the first target pushing subsequence according to the R sorting result.
Randomly selecting a target pushing sub-set from the M target pushing sub-sets, namely, taking the selected target pushing sub-set as a random variable, substituting the target pushing sub-set into an information entropy calculation formula, namely, calculating the information quantity contained in each piece of information, and putting the obtained conclusion into a data size comparison model for information entropy data sequencing to obtain a target pushing sub-sequence; and when the M-th information is selected, inputting the obtained information entropy data into a data size comparison model for training, and sequencing the information content size to obtain M target pushing subset sequences.
S300: retrieving a storage user set based on a push database, traversing the storage user set to extract data, and obtaining a user basic information set;
s400: extracting a user flow use set and a user use time window according to the user basic information set;
specifically, a storage user set is called through a push database, namely, an object set pushed in a historical time period is called from the push database, data extraction is carried out on the storage user set, namely, a process of obtaining target data from an information pool, information comprising the name of a user, the use time of the user, the use flow of the user, the access degree of the user to information and the like is obtained from the storage user set, and a user basic information set is formed according to the obtained information. The information of the client mobile phone APP can be called from the user basic information set, the time window can be obtained from the statistical record of the client mobile phone APP, and the user traffic use set and the user use time window, namely the traffic data use amount of the user and the distribution time period of the user using the mobile phone, are extracted according to the user basic information set.
Illustratively, the storage user set is called through the information in the push database, namely a push object set is obtained from the database; carrying out data extraction on the user set in the push database according to the push object set, wherein the data extraction comprises user names, user use flow, user information access degree and the like, so as to obtain a user basic information set; and obtaining a user flow using set and a set of mobile phone using distribution time periods of the user according to the information in the set, namely that the user uses the mobile phone in a certain time period in one day. By extracting and integrating the basic information of the user, a high-quality user is searched, so that the technical effect of improving the pushing efficiency can be achieved.
S500: acquiring a peak time window based on the user using time window, and extracting terminal calling tags from the user basic information set according to the peak time window to acquire N terminal calling tags;
specifically, a specific time is set, for example, a peak time window refers to a time period with the longest using time of a mobile phone in one day, the time window can be obtained from a statistical record of the APP of the mobile phone of the client, and the time period with the longest using time is obtained according to the distributed time period of the mobile phone; and carrying out terminal calling on the user basic information set according to the peak time window, namely obtaining a set of used APP in the period of time with the longest time of using the mobile phone through the user basic information set, classifying the APP in the period of time, and carrying out label extraction on the called APP, such as short video APP, long video APP, reading APP and the like, so as to obtain N terminal calling labels.
Further, the method comprises the following steps:
s510: terminal calling tag extraction is carried out on the user basic information set based on the peak time window, Q terminal calling tag sets are obtained, and the Q terminal calling tag sets are in one-to-one correspondence with the user basic information set;
s520: traversing the Q terminal call label sets to respectively construct call histograms to obtain Q terminal call histograms;
s530: and screening the Q terminal call histograms by using a preset time threshold, so as to obtain N terminal call labels.
Specifically, terminal call label extraction is performed on the user basic information set based on the peak time window, namely, in a period of longest use of a mobile phone in one day, a user classifies the used APP to obtain Q terminal call label sets, Q can be integers such as 1 and 2, wherein the Q terminal call label sets are in one-to-one correspondence with the user basic information sets, namely Q user basic information sets are provided, terminal call label extraction is performed on each user basic information set, one user information corresponds to one terminal call label set, call histograms are respectively constructed by searching the Q terminal call label sets, Q terminal call histograms are obtained, namely, in the period of use of time in the peak time window, the Q terminal call histograms are obtained by taking time as an X axis, and each label is taken as a Y axis; and screening the Q terminal call histograms by using a preset time threshold, wherein the preset time threshold is set by a worker according to actual conditions, the preset time threshold is not limited herein, the preset time threshold is used as a dividing standard, APP used in a time period exceeding the preset time threshold is worth analyzing, and the APP labels exceeding the preset time threshold can obtain N terminal call labels needing to be analyzed. Through analysis of the user basic information set, the short message pushing is more intelligent, and the requirements of people are met.
S600: screening the N terminal call labels according to a preset label set to obtain P adaptation call labels;
s700: carrying out serialization processing on the P adaptation call labels according to a user flow using set to obtain an adaptation call label sequence;
specifically, the preset label set is an APP label required by a worker, and is manually set according to the information type required to be pushed, including entertainment, short video and the like, and the N terminal call labels are screened according to the manually set label set, that is, among the N terminal call labels, the label which does not conform to the preset label set can be directly removed, so that P adaptation call labels are obtained, that is, the label which conforms to the preset label set. Firstly, sorting is carried out according to a user flow use set, wherein the more the flow is, the more the user flow package quantity is, the more the link is likely to be opened; in this range, the P adaptation call labels are serialized according to the user traffic usage set, so that one user traffic usage may correspond to X labels, and the adaptation call label sequences are obtained by sequencing the traffic usage sizes of the labels. By analyzing the user data, what kind of user is a potential target user can be obtained.
S800: inputting the M target pushing subsequences and the adaptive call tag sequences into an information pushing model to obtain an information pushing scheme;
specifically, the M target pushing sub-sequences refer to calculating information amount contained in information, sorting according to the information content, and inputting the information pushing sub-sequences into an information pushing model together with the adaptation calling label sequence, namely, the information pushing model constructed by a machine learning neural network algorithm, so as to obtain an information pushing scheme, namely, through analyzing data used by different information contents and different time slot flows, information pushing with different sizes is obtained, wherein the information pushing scheme is used for providing different time slots.
Further, the method comprises the following steps:
s810: constructing the information pushing model by taking a BP neural network as a basic framework, wherein input data of the information pushing model are the M target pushing subsequences and the adaptive calling tag sequences, and output data are an information pushing scheme;
s820: acquiring historical sample data, wherein the historical sample data comprises a plurality of sample target pushing subsequences, a sample adaptation calling tag sequence and a sample information pushing scheme;
s830: and dividing the historical sample data into a training set and a verification set, performing training supervision on the information pushing model by using the training set, and verifying the information pushing model after training by using the verification set until the information pushing model meets the preset condition, thereby obtaining the information pushing model after construction.
Specifically, the process of constructing the information push model is as follows: based on BP neural network in machine learning, constructing a network structure of the information pushing model, wherein the information pushing model comprises a plurality of simple units simulating human brain neurons, the information pushing model can form parameters such as weight, threshold and the like connected between the simple units in the supervision training process, and the information pushing model after training can carry out complex nonlinear logic operation according to input data to output a predicted information pushing scheme; the input data of the information pushing model is the M target pushing subsequences and the adaptation calling tag sequences, namely the result after the information content is ordered and the adaptation calling tag after the traffic is used for ordering, and the output data is the information pushing scheme, namely the information pushing is carried out in the time period. Acquiring historical sample data, wherein the historical sample data comprises a plurality of sample target pushing subsequences, sample adaptation calling tag sequences and a sample information pushing scheme, and the sample information pushing scheme refers to a past information pushing scheme; further carrying out data labeling on the historical sample data and dividing the historical sample data according to a certain proportion to obtain a training set, a verification set and a test set; inputting a plurality of sample data in a training set into an information pushing model, and performing supervision training on an information pushing model output scheme by using sample parameters by using a verification set to enable the information pushing model output scheme to be consistent with the sample information pushing scheme; after the data in the training set is trained, the accuracy test is carried out on the information Raisson model by utilizing the test set, a plurality of sample target pushing subsequences and sample adapting label sequences are respectively input into the model, a plurality of pushing schemes are obtained as actual schemes, a plurality of sample pushing schemes corresponding to the input data in the test set are used as expected outputs, the error between the actual outputs and the expected outputs is calculated, gradient descent updating is carried out on control parameters, in short, the error between the actual outputs and the expected outputs is used as a loss function, the smaller the loss function is, the smaller the description error is, the information pushing model with the accuracy meeting the preset conditions can be obtained, and the accuracy of the information pushing scheme of the information pushing model is improved. By constructing the information pushing model, the accuracy of the model is trained, the control accuracy of an information pushing scheme is improved, the phenomenon that pushing information is not intelligent enough is avoided, and the problem of low information pushing efficiency is solved.
Further, the method comprises the following steps:
s840: collecting click data of a user in the adaptation calling tag sequence;
s850: according to the click data, an invalid record of the storage user set is built;
s860: analyzing the invalid records and constructing insensitive key phrase of the storage user set;
s870: extracting key word groups in the M target push subsequences, and matching the key word groups with insensitive key word groups to obtain a matching result;
s880: and screening the adaptation calling tag sequence according to a matching result to obtain an optimized adaptation tag, and pushing information by using the optimized adaptation tag.
Specifically, according to a user basic information set, collecting click data of a user in the adaptation calling tag sequence to obtain whether the pushed information user clicks to see; according to the click data, information which is pushed but not clicked by a user is obtained, and an invalid record of the stored user set is built through the information; analyzing the invalid record, and constructing insensitive key word groups of the storage users, namely, users are not interested in the information, do not click in, and have poor pushing effects on the users, namely, the information can not be pushed to the users; extracting key phrases in the M target pushing subsequences, namely finding the key phrases from the M target pushing subsequences, namely classifying tags, and matching the key phrases with insensitive key phrases to obtain a matching result, namely obtaining information which is not interested by the user but needs pushing; and screening the obtained information on the adaptation calling tag sequence, namely deleting tags which are not interested by the user in the information to be pushed, obtaining an optimized adaptation tag, pushing the information according to the optimized adaptation tag, and carrying out feedback adjustment. The method is used for improving the quality of pushing users, improving the pushing efficiency and carrying out dimension reduction on the number of the users.
S900: and carrying out intelligent pushing of the short message according to the information pushing scheme.
Specifically, the M target pushing sub-sequences and the adaptive calling tag sequences are input into a trained information pushing model, an information pushing scheme is obtained according to user feedback, and short messages are pushed according to the information pushing scheme, namely, different types of short messages are recommended to a required user in different time periods. The intelligent short message pushing device solves the technical problems of low intelligent degree and poor pushing effect of short message pushing, and improves the pushing efficiency of information.
Example two
Based on the same inventive concept as the intelligent short message pushing method based on user analysis in the foregoing embodiment, as shown in fig. 4, the present application further provides a system based on the intelligent short message pushing method based on user analysis, where the system includes:
the target pushing information collection module 11 is used for obtaining a target pushing information collection, analyzing the target pushing information collection to obtain M target pushing sub-collections, wherein M is an integer greater than or equal to 1;
the target pushing subset processing module 12, where the target pushing subset processing module 12 is configured to perform information entropy calculation on the M target pushing subsets, and perform serialization processing on the M target pushing subsets according to a calculation result, to obtain M target pushing subsequences;
The user basic information collection module 13 is used for pushing a database to call a storage user collection, traversing the storage user collection to extract data, and obtaining a user basic information collection;
a user traffic and time usage module 14, where the user traffic and time usage module 14 is configured to extract a user traffic usage set and a user usage time window according to the user basic information set;
the terminal call acquisition module 15 is used for acquiring a peak time window by the user using the time window, and extracting terminal call labels from the user basic information set according to the peak time window to acquire N terminal call labels;
the terminal call tag screening module 16 is configured to screen the N terminal call tags according to a preset tag set, and obtain P adaptation call tags;
the adaptation call tag processing module 17 is configured to perform serialization processing on the P adaptation call tags according to a user traffic usage set, so as to obtain an adaptation call tag sequence;
The information pushing scheme obtaining module 18, where the information pushing scheme obtaining module 18 is configured to input the M target pushing subsequences and the adaptation calling tag sequences into an information pushing model to obtain an information pushing scheme;
the short message intelligent pushing module 19, the short message intelligent pushing module 19 is used for intelligent pushing of short messages according to the information pushing scheme.
Further, the embodiment of the application further comprises:
the historical push information collection module is used for extracting push information in the past time from the push database to obtain a historical push information collection;
the information type collection module is used for extracting information from the historical push information collection by taking the push information type as an index to obtain an information type collection;
the analysis decision tree constructing module is used for constructing an analysis decision tree according to the information type set and the historical push information set;
the target pushing subset obtaining module is used for inputting the target pushing information set into the analysis decision tree to obtain M target pushing subsets.
Further, the embodiment of the application further comprises:
the analysis decision tree root node building module is used for building a root node of the analysis decision tree according to the historical push information set;
the leaf node type marking module is used for traversing the information type set to construct a plurality of leaf nodes of the analysis decision tree and marking the leaf nodes in a type mode according to the information type;
the analysis set obtaining module is used for obtaining an analysis set by inputting the historical push information set into the analysis decision tree;
and the analysis decision tree construction completion module is used for matching the information type set with the analysis set, and obtaining the constructed analysis decision tree if the matching success rate exceeds a preset threshold.
Further, the embodiment of the application further comprises:
the characteristic information entropy collection module is used for traversing the M target pushing subsets to carry out information theory coding operation so as to obtain M characteristic information entropy collections;
the first target pushing sub-sequence module is used for randomly selecting a target pushing sub-set from the M target pushing sub-sets, taking the target pushing sub-set as a first target pushing sub-set, extracting a first characteristic information entropy set corresponding to the first target pushing sub-set, and training an input data size comparison model to obtain a first target pushing sub-sequence;
The second target pushing sub-sequence module is used for randomly selecting a target pushing sub-set from the M target pushing sub-sets, taking the target pushing sub-set as a second target pushing sub-set, extracting a second characteristic information entropy set corresponding to the second target pushing sub-set, and training an input data size comparison model to obtain a second target pushing sub-sequence;
the M-th target pushing sub-sequence module is used for randomly selecting a target pushing sub-set from the M target pushing sub-sets, taking the target pushing sub-set as the M-th target pushing sub-set, extracting an M-th characteristic information entropy set corresponding to the M-th target pushing sub-set, and training the input data size comparison module to obtain the M-th target pushing sub-sequence.
Further, the embodiment of the application further comprises:
the terminal calling tag set obtaining module is used for extracting terminal calling tags of the user basic information set through the peak time window to obtain Q terminal calling tag sets, wherein the Q terminal calling tag sets are in one-to-one correspondence with the user basic information set;
The call histogram constructing module is used for traversing the Q terminal call label sets to respectively construct call histograms to obtain Q terminal call histograms;
and the terminal call label obtaining module is used for screening the Q terminal call histograms by utilizing a preset time threshold value so as to obtain N terminal call labels.
Further, the embodiment of the application further comprises:
the information pushing model construction module is used for constructing the information pushing model by taking a BP neural network as a basic framework, input data of the information pushing model are the M target pushing subsequences and the adaptive calling tag sequences, and output data are an information pushing scheme;
the system comprises a historical sample data acquisition module, a data processing module and a data processing module, wherein the historical sample data acquisition module is used for acquiring historical sample data, and the historical sample data comprises a plurality of sample target pushing subsequences, sample adaptation calling tag sequences and a sample information pushing scheme;
the information pushing model construction completion module is used for dividing the historical sample data into a training set and a verification set, performing training supervision on the information pushing model by using the training set, and verifying the information pushing model after training by using the verification set until the information pushing model meets preset conditions, so as to obtain the information pushing model after construction.
Further, the embodiment of the application further comprises:
the user click data acquisition module is used for acquiring click data of a user in the adaptation calling tag sequence;
the invalid record constructing module is used for constructing the invalid record of the storage user set according to the click data;
the insensitive key phrase building module is used for analyzing the invalid records and building insensitive key phrases of the stored user set;
the keyword group matching module is used for extracting the keyword groups in the M target pushing subsequences and matching the keyword groups with insensitive keyword groups to obtain a matching result;
and the information pushing module is used for screening the adaptation calling tag sequence according to the matching result to obtain an optimized adaptation tag, and pushing information by using the optimized adaptation tag.
For a specific embodiment of a short message intelligent pushing system based on user analysis, reference may be made to the above embodiment of a short message intelligent pushing method based on user analysis, which is not described herein. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one embodiment, a computer device is provided, which may be a server, the internal structure of which may be as shown in fig. 4. The computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device is used for storing news data, time attenuation factors and other data. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by the processor to realize a short message intelligent pushing method based on user analysis.
Those skilled in the art will appreciate that the structures shown in FIG. 4 are block diagrams only and do not constitute a limitation of the computer device on which the present aspects apply, and that a particular computer device may include more or less components than those shown, or may combine some of the components, or have a different arrangement of components.
In one embodiment, a computer device is provided comprising a memory and a processor, the memory having stored therein a computer program, the processor when executing the computer program performing the steps of: acquiring a target pushing information set, and analyzing the target pushing information set to acquire M target pushing subsets, wherein M is an integer greater than or equal to 1; information entropy calculation is carried out on the M target pushing subsets, and serialization processing is carried out on the M target pushing subsets according to calculation results, so that M target pushing subsequences are obtained; retrieving a storage user set based on a push database, traversing the storage user set to extract data, and obtaining a user basic information set; extracting a user flow use set and a user use time window according to the user basic information set; acquiring a peak time window based on the user using time window, and extracting terminal calling tags from the user basic information set according to the peak time window to acquire N terminal calling tags; screening the N terminal call labels according to a preset label set to obtain P adaptation call labels; carrying out serialization processing on the P adaptation call labels according to a user flow using set to obtain an adaptation call label sequence; inputting the M target pushing subsequences and the adaptive call tag sequences into an information pushing model to obtain an information pushing scheme; and carrying out intelligent pushing of the short message according to the information pushing scheme.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples merely represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the invention. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.