CN113762298B - Similar crowd expansion method and device - Google Patents

Similar crowd expansion method and device Download PDF

Info

Publication number
CN113762298B
CN113762298B CN202010580213.1A CN202010580213A CN113762298B CN 113762298 B CN113762298 B CN 113762298B CN 202010580213 A CN202010580213 A CN 202010580213A CN 113762298 B CN113762298 B CN 113762298B
Authority
CN
China
Prior art keywords
model
user
data
classification
training
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010580213.1A
Other languages
Chinese (zh)
Other versions
CN113762298A (en
Inventor
谢宏斌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Jingdong Century Trading Co Ltd
Beijing Wodong Tianjun Information Technology Co Ltd
Original Assignee
Beijing Jingdong Century Trading Co Ltd
Beijing Wodong Tianjun Information Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Jingdong Century Trading Co Ltd, Beijing Wodong Tianjun Information Technology Co Ltd filed Critical Beijing Jingdong Century Trading Co Ltd
Priority to CN202010580213.1A priority Critical patent/CN113762298B/en
Publication of CN113762298A publication Critical patent/CN113762298A/en
Application granted granted Critical
Publication of CN113762298B publication Critical patent/CN113762298B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a similar crowd expansion method and device, and relates to the technical field of computers. One embodiment of the method comprises the following steps: acquiring a specific user corresponding to the target object, generating characteristic data of the specific user according to basic attribute data and behavior data of the specific user, wherein the specific user comprises: seed users and target users; training an algorithm model of an antagonistic network learning algorithm by adopting a random gradient method according to the characteristic data of a specific user; and determining the extensible user corresponding to the target object from the target user based on the algorithm model according to the characteristic data of the target user. The embodiment carries out semi-supervised training on the specific user based on the antagonistic network learning algorithm, avoids the sensitivity problem of the adaptive threshold selection of the unsupervised learning and the over-fitting problem of the supervised learning and special optimization method, and improves the accuracy of the trained algorithm model.

Description

Similar crowd expansion method and device
Technical Field
The invention relates to the technical field of computers, in particular to a similar crowd expansion method and device.
Background
The principle of similar crowd expansion is to find out the crowd similar to the seed users from mass target users based on some sub-users so as to expand the scale of the seed users. Currently, similar crowd expansion methods are mainly divided into three categories: the first is a clustering method of unsupervised learning, which distributes target users according to the class clusters to which the seed users belong; secondly, training a classification model by using seed users, and sorting according to the predicted value of the target user; and thirdly, a feature optimization method is adopted, similarity calculation is carried out on the target user according to the selected features, and then the target user is screened.
In the process of implementing the present invention, the inventor finds that at least the following problems exist in the prior art: 1. for the cluster method of unsupervised learning, it is difficult to determine the number of class clusters and the threshold value; 2. for the training method with supervised learning, the training method is easy to be overfitted to a seed user, so that generalization capability on a target user is poor; 3. for the feature-preferred method, it is difficult to screen out independent salient features and overfitting is easily caused.
Disclosure of Invention
In view of the above, the embodiment of the invention provides a similar crowd extension method and device, which can perform semi-supervised training on specific users based on an anti-network learning algorithm, avoid the problems of sensitivity of adaptive threshold selection of unsupervised learning and over-fitting of supervised learning and special optimization methods, and improve the accuracy of a trained algorithm model.
In order to achieve the above object, according to a first aspect of the embodiments of the present invention, there is provided a similar crowd expanding method.
The method for expanding the similar crowd comprises the following steps: acquiring a specific user corresponding to a target object, and generating characteristic data of the specific user according to basic attribute data and behavior data of the specific user, wherein the specific user comprises: seed users and target users; training an algorithm model of an antagonistic network learning algorithm by adopting a random gradient method according to the characteristic data of the specific user; and determining the extensible user corresponding to the target object from the target users according to the characteristic data of the target users based on the algorithm model.
Optionally, the training the algorithm model of the challenge network learning algorithm by adopting a random gradient method according to the characteristic data of the specific user includes: initializing a classification model of an countermeasure network learning algorithm; training a discrimination parameter according to the characteristic data of the specific user and the classification model by a random gradient ascending method to obtain a discrimination model corresponding to the discrimination parameter; training classification parameters by using a random gradient descent method and utilizing a loss function of the discrimination model, and updating the classification model according to the trained classification parameters; judging whether the judging model and the updated classifying model accord with preset conditions or not; if yes, determining the updated classification model as a target classification model, and determining the discrimination model as a target discrimination model; if not, carrying out model training by utilizing the characteristic data of the specific user and the updated classification model until the discrimination model and the classification model obtained by training meet preset conditions.
Optionally, training the discrimination parameters according to the feature data of the specific user and the classification model by using a random gradient ascent method to obtain a discrimination model corresponding to the discrimination parameters, including: sampling at least one first set of classified samples from the particular user; inputting the feature data corresponding to the at least one first classification sample set into the classification model to obtain a prediction classification result corresponding to the at least one first classification sample set; sampling at least one marked sample set from the seed user and the negative user, and determining a real classification result corresponding to the at least one marked sample set, wherein the real classification result corresponding to the seed user is 1, and the real classification result corresponding to the negative user is 0; constructing a discrimination training set by utilizing a prediction classification result corresponding to the at least one first classification sample set and a real classification result corresponding to the at least one mark sample set; training the discrimination parameters according to the discrimination training set by a random gradient rising method.
Optionally, the training the classification parameters by using the loss function of the discriminant model through a random gradient descent method includes: determining a loss function of the discrimination model; sampling at least one second set of classified samples from the particular user; and inputting the characteristic data corresponding to the at least one second classification sample set into a loss function of the discrimination model, and training the classification parameters by a random gradient descent method.
Optionally, the initializing a classification model of the challenge network learning algorithm includes: acquiring an initial value of a classification parameter, and initializing by directly utilizing the initial value of the classification parameter; and sampling at least one classification training set from the seed user and the negative user, and pre-training at least one initial classification model by utilizing the at least one classification training set to finish initialization of the classification model.
Optionally, the obtaining the target classification model and the target discrimination model of the countermeasure network learning algorithm includes: obtaining at least one optional classification model and at least one optional discriminant model corresponding to the at least one initial classification model; determining the target classification model and the target discrimination model from the at least one selectable classification model and the at least one selectable discrimination model.
Optionally, the method further comprises: randomly sampling the negative user from the target users; and sampling the negative user from the target user according to the seed user behavior data and the second behavior data of the seed user; the first behavior data of the negative user and the first behavior data of the seed user are intersected, and the second behavior data of the negative user and the second behavior data of the seed user are not intersected.
Optionally, the behavior data includes: first behavior data and second behavior data; and generating feature data of the specific user according to the basic attribute data and the behavior data of the specific user, including: acquiring the basic attribute data, the first behavior data and the second behavior data; processing the first line of data based on a preset embedded feature processing rule to generate embedded feature data corresponding to the first line of data; processing the second behavior data based on a preset word segmentation feature processing rule to generate word segmentation feature data corresponding to the second behavior data; and combining the basic attribute data, the embedded feature data and the word segmentation feature data to generate the feature data.
Optionally, the processing the first row of data based on the preset embedded feature processing rule, to generate embedded feature data corresponding to the first row of data, includes: acquiring at least one item attribute data corresponding to the first row of data; according to a preset time threshold corresponding to the at least one item attribute data, carrying out segmentation processing on the at least one item attribute data to obtain a behavior sequence corresponding to the at least one item attribute data; embedding the behavior sequence corresponding to the at least one item attribute data by utilizing a word vector embedding algorithm to obtain sub-embedded feature data corresponding to the at least one item attribute data; and combining the sub-embedded feature data corresponding to the at least one item attribute data to generate the embedded feature data corresponding to the first row of data.
Optionally, the processing the second behavior data based on the preset word segmentation feature processing rule, to generate word segmentation feature data corresponding to the second behavior data, includes: acquiring an article description sentence corresponding to the second behavior data; performing word segmentation processing on the article description sentence to obtain at least one word segment, and filtering and screening the at least one word segment; and generating word segmentation characteristic data corresponding to the second behavior data by utilizing the filtered word segmentation.
Optionally, the determining, based on the algorithm model, the extensible user corresponding to the target object from the target users according to the feature data of the target users includes: inputting the characteristic data of the target user into the target classification model to obtain a prediction classification result corresponding to the target user; and selecting the expandable user from the target users according to the prediction classification result corresponding to the target users based on preset expansion conditions.
In order to achieve the above object, according to a second aspect of the embodiments of the present invention, there is provided a similar crowd expanding device.
The embodiment of the invention provides a similar crowd expanding device, which comprises: the generating module is used for acquiring a specific user corresponding to the target object, generating characteristic data of the specific user according to basic attribute data and behavior data of the specific user, wherein the specific user comprises: seed users and target users; the training module is used for training an algorithm model of an antagonistic network learning algorithm by adopting a random gradient method according to the characteristic data of the specific user; and the determining module is used for determining the extensible user corresponding to the target object from the target users according to the characteristic data of the target users based on the algorithm model.
Optionally, the training module is further configured to: initializing a classification model of an countermeasure network learning algorithm; training a discrimination parameter according to the characteristic data of the specific user and the classification model by a random gradient ascending method to obtain a discrimination model corresponding to the discrimination parameter; training classification parameters by using a random gradient descent method and utilizing a loss function of the discrimination model, and updating the classification model according to the trained classification parameters; judging whether the judging model and the updated classifying model accord with preset conditions or not; if yes, determining the updated classification model as a target classification model, and determining the discrimination model as a target discrimination model; if not, carrying out model training by utilizing the characteristic data of the specific user and the updated classification model until the discrimination model and the classification model obtained by training meet preset conditions.
Optionally, the training module is further configured to: sampling at least one first set of classified samples from the particular user; inputting the feature data corresponding to the at least one first classification sample set into the classification model to obtain a prediction classification result corresponding to the at least one first classification sample set; sampling at least one marked sample set from the seed user and the negative user, and determining a real classification result corresponding to the at least one marked sample set, wherein the real classification result corresponding to the seed user is 1, and the real classification result corresponding to the negative user is 0; constructing a discrimination training set by utilizing a prediction classification result corresponding to the at least one first classification sample set and a real classification result corresponding to the at least one mark sample set; training the discrimination parameters according to the discrimination training set by a random gradient rising method.
Optionally, the training module is further configured to: determining a loss function of the discrimination model; sampling at least one second set of classified samples from the particular user; and inputting the characteristic data corresponding to the at least one second classification sample set into a loss function of the discrimination model, and training the classification parameters by a random gradient descent method.
Optionally, the training module is further configured to: acquiring an initial value of a classification parameter, and initializing by directly utilizing the initial value of the classification parameter; and sampling at least one classification training set from the seed user and the negative user, and pre-training at least one initial classification model by utilizing the at least one classification training set to finish initialization of the classification model.
Optionally, the training module is further configured to: obtaining at least one optional classification model and at least one optional discriminant model corresponding to the at least one initial classification model; determining the target classification model and the target discrimination model from the at least one selectable classification model and the at least one selectable discrimination model.
Optionally, the apparatus further comprises a sampling module for: randomly sampling the negative user from the target users; and sampling the negative user from the target user according to the seed user behavior data and the second behavior data of the seed user; the first behavior data of the negative user and the first behavior data of the seed user are intersected, and the second behavior data of the negative user and the second behavior data of the seed user are not intersected.
Optionally, the behavior data includes: first behavior data and second behavior data; the generation module is further configured to: acquiring the basic attribute data, the first behavior data and the second behavior data; processing the first line of data based on a preset embedded feature processing rule to generate embedded feature data corresponding to the first line of data; processing the second behavior data based on a preset word segmentation feature processing rule to generate word segmentation feature data corresponding to the second behavior data; and combining the basic attribute data, the embedded feature data and the word segmentation feature data to generate the feature data.
Optionally, the generating module is further configured to: acquiring at least one item attribute data corresponding to the first row of data; according to a preset time threshold corresponding to the at least one item attribute data, carrying out segmentation processing on the at least one item attribute data to obtain a behavior sequence corresponding to the at least one item attribute data; embedding the behavior sequence corresponding to the at least one item attribute data by utilizing a word vector embedding algorithm to obtain sub-embedded feature data corresponding to the at least one item attribute data; and combining the sub-embedded feature data corresponding to the at least one item attribute data to generate the embedded feature data corresponding to the first row of data.
Optionally, the generating module is further configured to: acquiring an article description sentence corresponding to the second behavior data; performing word segmentation processing on the article description sentence to obtain at least one word segment, and filtering and screening the at least one word segment; and generating word segmentation characteristic data corresponding to the second behavior data by utilizing the filtered word segmentation.
Optionally, the determining module is further configured to: inputting the characteristic data of the target user into the target classification model to obtain a prediction classification result corresponding to the target user; and selecting the expandable user from the target users according to the prediction classification result corresponding to the target users based on preset expansion conditions.
To achieve the above object, according to a third aspect of the embodiments of the present invention, there is provided an electronic apparatus.
An electronic device of an embodiment of the present invention includes: one or more processors; and the storage device is used for storing one or more programs, and when the one or more programs are executed by one or more processors, the one or more processors realize the similar crowd extension method of the embodiment of the invention.
To achieve the above object, according to a fourth aspect of the embodiments of the present invention, there is provided a computer-readable medium.
A computer readable medium of an embodiment of the present invention has a computer program stored thereon, which when executed by a processor implements the similar crowd extension method of the embodiment of the present invention.
One embodiment of the above invention has the following advantages or benefits: the basic attribute data and the behavior data of the specific user can be utilized to generate the characteristic data, and the behavior data is introduced, so that the method has important significance for improving the conversion rate, and the accuracy of a trained algorithm model can be improved; then training an algorithm model of an anti-network learning algorithm by combining characteristic data of a specific user through a random gradient method, and performing semi-supervised training on the specific user based on the anti-network learning algorithm, wherein the method is different from a method for performing supervised training by only using seed users and a method based on unsupervised clustering by all users, so that the adaptive threshold selection sensitivity problem of unsupervised learning and the overfitting problem of supervised learning and a special optimization method are avoided; and finally, selecting an extensible user from target users by using a trained algorithm model, and completing similar crowd extension.
Further effects of the above-described non-conventional alternatives are described below in connection with the embodiments.
Drawings
The drawings are included to provide a better understanding of the invention and are not to be construed as unduly limiting the invention. Wherein:
FIG. 1 is a schematic diagram of the main steps of a similar crowd expansion method according to an embodiment of the invention;
FIG. 2 is a schematic illustration of at least one item attribute data acquired in accordance with an embodiment of the present invention;
FIG. 3 is a schematic diagram of the main flow of a method of generating feature data for a particular user according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of the main flow of a method of training a classification model and a discriminant model according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a similar crowd expansion system according to an embodiment of the invention;
FIG. 6 is a schematic diagram of the main modules of a similar crowd expansion device according to an embodiment of the invention;
FIG. 7 is an exemplary system architecture diagram in which embodiments of the present invention may be applied;
Fig. 8 is a schematic diagram of a computer system suitable for use in implementing an embodiment of the invention.
Detailed Description
Exemplary embodiments of the present invention will now be described with reference to the accompanying drawings, in which various details of the embodiments of the present invention are included to facilitate understanding, and are to be considered merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the invention. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
The expansion of similar people needs to solve two main problems, namely how to determine seed users or how to determine the characteristics of the seed users according to the seed users; and secondly, expanding the similarity according to the determined seed users and the characteristics thereof. The seed user is a user interested in something or a store, for example, the user purchases something or collects something or pays attention to the store, and the specific source may be provided by a specific operator. The similar crowd expansion is to determine the characteristics of the seed users, and then the crowd similar to the characteristics of the seed users can be mined according to the characteristics of the seed users, and the mined crowd is considered to be like the seed users and interested in something or a store. Investigation shows that the existing similar crowd expansion methods are mainly divided into the following three types:
1. And clustering the seed users and the target users based on an unsupervised learning clustering method, and distributing the target users according to the class clusters to which the seed users belong. Unsupervised learning belongs to one of the machine learning algorithms, and is used to process sample sets that are not labeled by classification when designing a classifier. Therefore, in the clustering algorithm of the unsupervised learning algorithm, the sample data category is unknown, and the sample set needs to be classified or clustered according to the similarity among samples, so as to try to minimize the intra-class gap and maximize the inter-class gap. In practical application, the labels of the samples cannot be known in advance under many conditions, that is, the categories corresponding to the samples are not trained, so that the classifier design can only be learned from the sample set without the sample labels. The target users refer to users other than seed users, and the purpose of the similar crowd expansion is to select users possibly interested in something or a store from among the target users, so that recommendation messages can be sent to the users.
For the clustering method based on the unsupervised learning, it is difficult to determine the number of class clusters, i.e. it is difficult to define in advance how many classes the seed users can be divided into, in practice this strong a priori knowledge guidance is lacking, and even if the number of class clusters can be determined, it is difficult to select an adaptability threshold to screen the users.
2. Based on a training method of supervised learning, a seed user is regarded as a positive sample, then a negative sample is acquired, classification model training is carried out, and finally sorting is carried out according to a predicted value of a target user. Supervised learning is one of the machine learning algorithms, which is the machine learning task that extrapolates a function from a labeled training dataset. In the supervised learning training method, a function (model parameter) is learned from a given training set, and when new data arrives, the result can be predicted according to the function. The training set includes positive samples and negative samples, the positive samples refer to samples belonging to a certain category, the negative samples refer to samples not belonging to a certain category, for example, in the process of image recognition of the letter A, the samples of the letter A belong to the positive samples, and the samples not belonging to the letter A belong to the negative samples. The supervised learning is the most common classification (different from clustering), an optimal model (the model belongs to a set of a certain function and is optimal under a certain evaluation criterion) is obtained by training the existing training samples (namely known data and corresponding output thereof), all the inputs are mapped into corresponding output by using the model, and the output is simply judged so as to realize the purpose of classification.
For training methods based on supervised learning, when the number of seed users is too small and the feature dimension is too large, the training method is easy to overfit to the seed users, so that generalization capability on target users is poor. Overfitting is the fact that the built machine learning model or the deep learning model performs too well in training samples, resulting in poor performance in the validation data set as well as in the test data set. Such as identifying a model of a dog, which requires training. Just, all training pictures in the training set are two-ha, so after multiple iterative training, the model is trained and performs well in the training set. However, placing a test sample of gold wool into the model that identifies the dog has a high probability that the model will output the result that the gold wool is not a dog. This results in a model overfitting, which, while performing well on the training set, performs exactly the opposite in the test set, being too covariance from a performance standpoint, as well as a loss function on the test set. Generalization capability refers to the ability of a learned model to predict unknown data.
3. Based on the feature optimization method, user features are optimized firstly, then seed user overall similarity calculation is carried out on the target user according to the selected features, and then the target user is screened.
For the feature-based optimization method, the relative independence among the features of the user needs to be assumed, and because the local features often have strong or weak dependency relationships, the independent significant features are difficult to screen, and the strong independent features are used, so that the overfitting phenomenon is easy to occur.
In order to solve the above problems, the embodiment of the present invention proposes a semi-supervised user classification scheme for countermeasure network learning, which is different from a method for performing supervised training by using only seed users, and is different from a method for performing supervised clustering based on all users, and the embodiment of the present invention performs supervised training on all users based on countermeasure network learning algorithm. The semi-supervised learning mechanism is introduced to solve the problems of adaptability threshold selection sensitivity of the non-supervised learning, supervised learning and overfitting of a special optimization method, wherein the semi-supervised learning is a learning method combining the supervised learning and the non-supervised learning, and mainly considers the problems of training and classifying by using a small amount of marked samples and a large amount of unmarked samples. In the semi-supervised learning of the embodiments of the present invention, the user classifier is learned with an antagonism network learning algorithm, the antagonism is aimed at learning the joint distribution of a sample set that includes both seed users and encompasses target users, that is, training using all data, not just seed users. Fig. 1 is a schematic diagram of main steps of a method for expanding a similar crowd according to an embodiment of the invention, and as shown in fig. 1, the main steps of the method for expanding a similar crowd may include steps S101 to S103.
Step S101: and acquiring the specific user corresponding to the target object, and generating characteristic data of the specific user according to the basic attribute data and the behavior data of the specific user.
The target object may be an article or store that is interested by the seed user, for example, if a user purchases an article W or pays attention to a store D, the user may be a seed user, and the article W or store D may be a target object; the specific user may include: a seed user and a target user, considering that the seed user is a user interested in something or a store, the target user refers to a user other than the seed user, so that a specific user is equivalent to all users, such as on a certain e-commerce platform, and then a specific user refers to a user registered on the e-commerce platform. Alternatively, the target object is a game, the game has an age limit, and then the specific users refer to all users meeting the age requirement. In the embodiment of the invention, the specific user can be determined according to the actual situation.
After the specific user is acquired, the characteristic data of the specific user can be generated according to the basic attribute data and the behavior data of the specific user. The basic attribute data may include user attributes, such as age, gender, school, height, marital, occupation, income, etc., behavior trends, such as liveness, sleep-in period, shopping places and time periods, return trends, etc., interest preferences, such as brands, categories, etc., which are often browsed, etc., which can reflect the characteristics of the user that are relatively stable over a period of time. In the embodiment of the invention, besides considering the long-term characteristics, the recent characteristics of the user, namely the latest user behavior data, are required to be introduced, wherein the user behavior data represent clicking, collecting, focusing and purchasing behaviors which are shown in a login session period. In the method for generating the characteristic data, the user behavior data is introduced, so that the method has important significance for improving the conversion rate, and in an e-commerce environment, the conversion rate can refer to the proportion of the purchasing behavior of the user after browsing.
Step S102: according to the characteristic data of the specific user, training an algorithm model of the countermeasure network learning algorithm by adopting a random gradient method.
After the characteristic data of the specific user is generated, an algorithm model of the countermeasure network learning algorithm can be trained by using a random gradient method by utilizing the generated characteristic data. The challenge network learning algorithm belongs to a deep learning model, and in the embodiment of the invention, the algorithm model of the challenge network learning algorithm may include: classification models and discriminant models. The classification model is mainly used for learning how to identify a target user as a potential extensible user, namely, the classification model is used for predicting whether the user has preference on a target object, so that a prediction result of the classification model is more accurate, and the discrimination model is cheated. The discrimination model needs to discriminate the true or false of the received sample. For example, given a training set that contains positive and negative samples, the goal is to learn according to what rules such training set is constructed. In the whole process, the classification model makes the prediction result more accurate in an effort, the discrimination model identifies the true and false of the sample in an effort, and the classification model and the discrimination model are constantly opposed with the lapse of time, so that the two models reach a dynamic balance finally: the classification model predicts whether the user has a preferential result on the target object more accurately, and the discrimination model can not identify the true and false samples. That is, the antagonism balance of the classification model and the discriminant model enables the classification model to learn the joint distribution from the seed user and the target user.
In the semi-supervised learning of the embodiments of the present invention, user classification is learned using an antagonistic network learning algorithm, the objective of antagonism is to learn the joint distribution of a sample set that includes both seed users and encompasses target users, that is, training is performed using all data, not just seed users. Therefore, training is performed using the feature data of the feature user in step S102.
In addition, the embodiment of the invention adopts a random gradient method to carry out model training. The stochastic gradient method is an optimization algorithm commonly used in machine learning and artificial intelligence to recursively approximate minimum deviation models and solve maxima or minima along the gradient ascending or descending direction. The random gradient method is to extract a batch of training data from a training set randomly to input and calculate the average gradient, and the method has the advantages of reducing the calculation cost of each iteration and accelerating the model training speed.
Step S103: and determining the extensible user corresponding to the target object from the target user based on the algorithm model according to the characteristic data of the target user.
Through training the algorithm model in step S102, a target classification model and a target discrimination model can be obtained, that is, in the model training process, if the classification model and the discrimination model reach balance, training can be stopped, and the classification model and the discrimination model at this time are the target classification model and the target discrimination model, and whether the target user has a preference for the target object can be predicted by using the target classification model, so that the extensible user can be selected from the target users.
Thus, as a referenceable embodiment of the present invention, determining, based on an algorithm model, an extensible user corresponding to a target object from among target users according to feature data of the target users may include: inputting the characteristic data of the target user into a target classification model to obtain a prediction classification result corresponding to the target user; based on preset expansion conditions, expandable users are selected from target users according to prediction classification results corresponding to the target users.
The preset expansion condition may be a preset classification result threshold, feature data of a certain target user M is input into the target classification model, and if the obtained prediction classification result is greater than the preset classification result threshold, the target user M is indicated to be an expandable user. The preset expansion condition can also be the preset number N of expansion users, the characteristic data of all target users are input into the target classification model, the obtained prediction classification results are ordered, and the target users with the top N rank are selected as expandable users. Of course, the preset expansion condition may also be other forms, and the embodiment of the present invention is not limited to this according to the actual situation setting.
According to the similar crowd expansion method provided by the embodiment of the invention, the characteristic data can be generated by utilizing the basic attribute data and the behavior data of the specific user, and the behavior data is introduced, so that the method has important significance for improving the conversion rate, and the accuracy of a trained algorithm model can be improved; then training an algorithm model of an anti-network learning algorithm by combining characteristic data of a specific user through a random gradient method, and performing semi-supervised training on the specific user based on the anti-network learning algorithm, wherein the method is different from a method for performing supervised training by only using seed users and a method based on unsupervised clustering by all users, so that the adaptive threshold selection sensitivity problem of unsupervised learning and the overfitting problem of supervised learning and a special optimization method are avoided; and finally, selecting an extensible user from target users by using a trained algorithm model, and completing similar crowd extension.
The similar crowd expansion function is located on the user portrait platform, and needs to rely on the user portrait system to acquire the user basic attributes as input, which describe the characteristics of user attributes, behavior trends, interest preferences and the like, reflect the characteristics of the user that are relatively stable in a certain period, and have been illustrated in step S101, which is not described here again. In addition to these stability features, it is also necessary to introduce user recent features, i.e. recent user behavior, which is of great importance for conversion rate improvement. However, it should be noted that the user behavior features are generally sparse, and the model training is performed directly by using the sparse features, which has the consequence of being easy to be over-fitted, so that the user behavior features need to be processed. Therefore, as a referenceable example of the present invention, generating feature data of a particular user from base attribute data and behavior data of the particular user includes:
step S1011, acquiring basic attribute data, first behavior data and second behavior data;
Step S1012, processing the first line of data based on a preset embedded feature processing rule to generate embedded feature data corresponding to the first line of data;
Step S1013, processing the second behavior data based on a preset word segmentation feature processing rule to generate word segmentation feature data corresponding to the second behavior data;
step S1014, combining the basic attribute data, the embedded feature data, and the word segmentation feature data to generate feature data.
The behaviors of the user may include a normal clicking behavior, a focusing behavior, a collecting behavior, a purchasing behavior and the like, and since the number of the normal clicking behaviors is relatively large and other behaviors are relatively small, the behaviors are separately processed in the embodiment of the invention, and the behavior data are divided into first behavior data and second behavior data, wherein the first behavior data is equivalent to the normal clicking behavior data with a large number, and the second behavior data is equivalent to other behavior data with a relatively small number, such as the focusing behavior, the collecting behavior, the purchasing behavior and the purchasing behavior.
Considering that the common clicking behaviors of the user reflect continuous changes of the user demands, the object sequences corresponding to the common clicking behaviors have strong correlation, and the objects can reflect the intrinsic demands of the user. However, since the common clicking behaviors of the users are scattered, the common clicking behaviors belong to sparse features, and the data are scattered in a high-dimensional space, if the common clicking behaviors of the users are directly used for forming feature data, the problem of overfitting of a trained model can be caused. Therefore, in the embodiment of the invention, the low-dimensional feature embedding processing is carried out on the common clicking behaviors of the user. The benefit of embedding the resulting user features is that global abstract features are obtained, so that the generalization is strong, and by way of illustration, a user wants to buy fruit, she looks at dragon fruit, orange, mango, and grapefruit, and finally buys the grapefruit, if the user is described by the focused, purchased, and purchased article word segmentation features, the user may like the grapefruit, thus resulting in an overfitting behavior, the user buying the grapefruit belongs to two distinct groups with the user buying the orange, and the user buying the grapefruit, but the fact is that the user buying the grapefruit has a preference for the orange, and even the mango and the dragon fruit.
In addition, the embedded low-dimensional features are more descriptive of inherent correlation properties in the item space than the sparse features. This is because for high dimensional vectors, such as a (1, 0), B (0, 1, 0), and C (0, 1), the distances between the three are equal, it is difficult to find whether a and B are similar or a and C are similar, but because the low dimensional space is dense, embedding more easily results in a and B or C being more similar (closer distance). Therefore, as a referenceable embodiment of the present invention, processing the first row of data based on a preset embedded feature processing rule, generating embedded feature data corresponding to the first row of data may include:
step S10121, at least one item attribute data corresponding to the first row of data is obtained;
Step S10122, according to a preset time threshold corresponding to at least one item attribute data, carrying out segmentation processing on the at least one item attribute data to obtain a behavior sequence corresponding to the at least one item attribute data;
step S10123, performing embedding processing on the behavior sequence corresponding to the at least one item attribute data by using a word vector embedding algorithm to obtain sub-embedded feature data corresponding to the at least one item attribute data;
Step S10124, combining the sub-embedded feature data corresponding to the at least one item attribute data to generate embedded feature data corresponding to the first row of data.
Wherein the item attribute data may include: the name of the article, the brand to which the article belongs, the class to which the article belongs, and the store of the article. After the first behavior data of the user is obtained, the article sequence corresponding to the first behavior data can be obtained, for example, the first behavior data of a user clicking 100 articles is obtained, and then the 100 article sequences corresponding to the first behavior data can be obtained. Since it is a fine-grained description of the user's needs, it is not desirable to use high-dimensional item title word as a feature vector, and therefore attributes of the item such as brands, categories, stores are used as alternative descriptions of the item. Therefore, in step S10121, at least one item attribute data corresponding to the first row of data is acquired, and fig. 2 is a schematic diagram of the acquired at least one item attribute data according to an embodiment of the present invention. In fig. 2, the first behavior data of the user is behavior data generated by clicking in the order from item a to item F, and the obtained item attribute data is 4 sequences, which are a sequence from item a to item F, a sequence from brand corresponding to item a to brand corresponding to item F, a sequence from category corresponding to item a to category corresponding to item F, and a sequence from store corresponding to item a to store corresponding to item F, respectively.
Because the depicting granularity of brands, categories and shops is larger, the situation that repeated elements appear in one sequence in a clicking sequence of a user is caused, and dense embedding is unfavorable, and therefore, adjacent repeated brands, categories and shops in at least one item attribute data are required to be subjected to segmented compression processing, and then the segmented compressed sequence is used for brands, categories and shops embedding, the scheme is more important than that of directly using 0-1 vectorization coding, and the scheme can make a final algorithm model have robustness. Firstly, dividing clicking sequence segments, namely obtaining at least one behavior sequence corresponding to the object attribute data, wherein different time thresholds are required to be set for constructing the corresponding behavior sequences due to different granularity of objects, brands, stores and categories. For a behavior sequence corresponding to the article, if the interval between two adjacent clicking time periods exceeds 1 minute, segmenting; for the behavior sequences corresponding to brands and shops, if the interval between two adjacent clicking time lengths exceeds 5 minutes, segmenting; and (3) segmenting the behavior sequence corresponding to the category if the interval between two adjacent clicking time periods exceeds 30 minutes. In addition, if the behavior sequence obtained by segmentation processing according to the preset time threshold still has adjacent repeated elements, one of the repeated elements is reserved.
After the behavior sequence corresponding to the at least one item attribute data is obtained, embedding processing can be performed on the behavior sequence corresponding to the at least one item attribute data by utilizing a word vector embedding algorithm, so as to obtain sub-embedded feature data corresponding to the at least one item attribute data. The word vector embedding algorithm is an algorithm for converting words into vectors, namely, the words in a dictionary are marked by vectors with certain dimensions.
In summary, the processing of the normal clicking behavior of the user obtains the embedded feature data of the user, which specifically includes: processing from four angles of clicked articles, brands, categories and shops, generating four corresponding sequences aiming at first behavior data to obtain four corresponding embedded feature vectors, namely four sub-embedded feature data, and splicing the four sub-embedded feature data to obtain embedded feature data of a user.
In the embodiment of the invention, the first behavior data is equivalent to the large number of common click behavior data, and the second behavior data is equivalent to the relatively small number of other behavior data, such as attention behavior, collection behavior, purchasing behavior and purchasing behavior, so that the first behavior data is subjected to embedded feature processing, and the second behavior data is subjected to word segmentation feature processing. As a referenceable embodiment of the present invention, processing the second behavior data based on a preset word segmentation feature processing rule, generating word segmentation feature data corresponding to the second behavior data may include:
step S10131, obtaining an article description sentence corresponding to the second behavior data;
step S10132, performing word segmentation processing on the object description sentence to obtain at least one segmented word, and filtering and screening the at least one segmented word;
step S10133, generating word segmentation feature data corresponding to the second behavior data by utilizing the filtered word segmentation.
The word segmentation statistical feature is that the attention, collection, purchase and purchase behaviors of a user are processed, namely, second behavior data of the user are processed, articles corresponding to the behaviors and article description sentences are obtained, then word segmentation processing is carried out on the article description sentences to obtain a plurality of segmented words, then filtering and screening are carried out on all the obtained segmented words, and finally the generated article title segmented word feature is used as a part of user feature data.
After processing the first row of data in step S1012 to generate embedded feature data corresponding to the first row of data, and processing the second row of data in step S1013 to generate word segmentation feature data corresponding to the second row of data, the base attribute data, the embedded feature data, and the word segmentation feature data may be combined to generate feature data of the user.
Fig. 3 is a schematic diagram of the main flow of a method of generating characteristic data of a specific user according to an embodiment of the present invention. As shown in fig. 3, the main flow of the method of generating feature data of a specific user may include:
Step S301, basic attribute data, first behavior data and second behavior data of a specific user are obtained;
Step S302, at least one item attribute data corresponding to the first row of data is obtained;
Step S303, carrying out segmentation processing on at least one item attribute data according to a preset time threshold corresponding to the at least one item attribute data to obtain a behavior sequence corresponding to the at least one item attribute data;
step S304, embedding the behavior sequence corresponding to the at least one item attribute data by utilizing a word vector embedding algorithm to obtain sub-embedded feature data corresponding to the at least one item attribute data;
Step S305, combining the sub-embedded feature data corresponding to at least one item attribute data to generate embedded feature data corresponding to the first row of data;
step S306, acquiring an article description sentence corresponding to the second behavior data;
Step S307, performing word segmentation processing on the object description sentence to obtain at least one word segment, and filtering and screening the at least one word segment;
Step S308, generating word segmentation feature data corresponding to the second behavior data by utilizing the filtered word segmentation;
Step S309, combining the basic attribute data, the embedded feature data and the word segmentation feature data to generate feature data of the specific user.
It should be noted that, in steps S302 to S305, the first behavior data is processed to generate the embedded feature data, and in steps S306 to S308, the second behavior data is processed to generate the word segmentation feature data, and the specific execution sequence may be adjusted according to the actual situation, so that the embedded feature data may be generated first, the word segmentation feature data may be regenerated, and the embedded feature data and the word segmentation feature data may also be generated simultaneously.
In the method for generating the feature data of the specific user, the long-term feature and the recent feature of the user can be comprehensively considered from three aspects of basic attribute data, first behavior data and second behavior data, so that the accuracy of the feature data and an algorithm model can be improved, the accuracy of the obtained expandable user can be further ensured, and in addition, the introduction of the feature of the user has important significance for improving the conversion rate. And taking the characteristics of the first behavior data, such as a large quantity and sparse characteristics into consideration, adopting an embedded characteristic processing method to obtain corresponding embedded characteristic data, and avoiding the problem of overfitting of a trained model.
The algorithm model for training the countermeasure network learning algorithm is an important component of the similar crowd expansion method of the embodiment of the invention. As a referenceable embodiment of the present invention, training an algorithm model against a network learning algorithm using a random gradient method according to characteristic data of a specific user may include: initializing a classification model of an countermeasure network learning algorithm; training the discrimination parameters according to the characteristic data and the classification model of the specific user by a random gradient ascending method to obtain a discrimination model corresponding to the discrimination parameters; training classification parameters by using a loss function of the discrimination model through a random gradient descent method, and updating the classification model according to the trained classification parameters; judging whether the judging model and the updated classifying model accord with preset conditions or not; if yes, determining the updated classification model as a target classification model, and determining the discrimination model as a target discrimination model; if not, carrying out model training by utilizing the characteristic data of the specific user and the updated classification model until the discrimination model and the classification model obtained by training meet the preset conditions.
The input of the classification model C (x) is the characteristic data of the user x, and the prediction classification result corresponding to the user x is outputFor the classification model C (x), the user x may be selected from a classification sample set composed of the seed user and the target user, that is, the classification model C (x) may predict classification results of the seed user and the target user. In the embodiment of the invention, a deep neural network can be used as a classification model for training, the number of neurons of an input layer is kept to be the same as the number of user characteristics, a plurality of layers of fully connected units are arranged in the middle, the activation function in the network selects the leak-Relu, batch regularization is carried out first, the number of neurons of the output layer is 2, and the softmax is used for predicting the probability distribution/>, of the user
It should be noted that for a seed user, the corresponding true classification result y=1, that is, the seed user may be regarded as a marked sample, the target user may be regarded as a sample to be marked, and the seed user belongs to a positive sample, and the negative sample selection problem, i.e., how to determine the user with y=0, will be discussed in detail. The negative sample is important for training the model, and how to select the high-quality negative sample is directly related to the intensity of the generalization capability of the model. In the embodiment of the invention, the negative user can be randomly sampled from the target users, namely, the negative user is randomly selected from the target users, and then the negative user is determined to be a marked sample, and the corresponding real classification result y=0, so that the negative user can be defined as the user with the real classification result of 0.
In addition, in the embodiment of the invention, the negative user can be sampled from the target user according to the seed user behavior data and the second behavior data of the seed user. The first behavior data of the negative user and the first behavior data of the seed user are intersected, and the second behavior data of the negative user and the second behavior data of the seed user are not intersected. The feature dimension is enlarged, potentially expanding the coverage, considering that the first behavior data of the user is introduced in the method of generating the feature data. Therefore, in order to enhance the generalization performance of the model, besides random sampling, a user having an intersection with the first behavior data of the seed user and a user having no intersection with the second behavior data of the seed user may be selected as a negative user, for example, the selected negative user has no intersection with the seed user in regard to attention, collection, purchase behavior, but has an intersection with the seed user in general click behavior. For example, the seed user clicks on pineapple, banana, cantaloupe, tomato, beef, egg, and finally purchases pineapple, tomato, beef; then when a negative user is selected, a user who purchased eggs, bananas, cantaloupe, but clicked on pineapple, tomatoes, beef can be selected. According to the behavior data of the seed user and the second behavior data of the seed user, the negative user is sampled from the target user, so that the seed user and the negative user are not differentiated in two stages in the feature space, and the enhancement of the generalization capability of the model is important.
Typically, the positive and negative sample ratio will be kept at 1:3, namely if the number of the seed users is 100, 300 users can be selected from the target users as negative users at random, or 300 users can be selected from the target users as negative users according to the seed user behavior data and the second behavior data of the seed users.
For the discriminant model D, a sample is givenThe discrimination model D can discriminate it as a counterfeit sample, that is, input the feature data of the user x and the prediction classification result/>, by the classification modelThe discrimination model D can determine the sampleIn order to forge the sample, namely, the judging model D can judge that the input classification result is the predicted classification result of the classification model and is not the real classification result. Similarly, for the discrimination model D, given one sample (x, y), the discrimination model D can distinguish the sample as a true sample, that is, the feature data of the input user x and the true classification result y, and the discrimination model D can determine that the sample (x, y) is a true sample, that is, the discrimination model D can determine that the input classification result is a true classification result, and can determine that the input user is a seed user or a negative user. It can be seen that the classification model C (x) corresponds to a user marking model for marking users in the set of classified samples, the real samples belonging to the marked samples. Also, sample/>The corresponding user is selected from a classified sample set consisting of seed user and target user, and the corresponding user of sample (x, y) is selected from a labeled sample set consisting of seed user and negative user, thus sample/>The users corresponding to the samples (x, y) may be the same user.
In the embodiment of the invention, the number of nodes of the input layer of the judging model D can be more than 2 neurons than the number of nodes of the classifying model C (x), the probability of a seed user and a non-seed user is represented respectively, the output layer is 1 node, logic Style regression is carried out, the judging probability of a true sample is output, the activating function in a network is also selected from the group consisting of the selection of the activating function in the network, batch regularization is carried out, the number of the neurons of the output layer is 2, and the final loss function is as follows:
Wherein D represents a discrimination model, and C represents a classification model; v (D, C) represents a cost function, which is a very small maximum function that can be decomposed into:
f (C) =max DV(D,C),minCmaxDV(D,C);Pseed(x)∪Pneg (x) represents the built-in distribution P ((x seed,y＝1)∪(xneg, y=0)) of the seed user and the negative user, the sample generated by this distribution is a real sample, log D (x, y) represents the likelihood that the discrimination model D recognizes the real sample; since the seed user is a marked user, P all (x) represents the target user distribution Representing the predicted probability distribution of the classification model for the target user as a potentially scalable user,Indicating the likelihood that the discrimination pattern identifies a counterfeit sample.
V (D, C) finally represents the expected likelihood, the training target is given a classification model C, a discrimination model D is obtained through maximum likelihood optimization V (D, C), namely, the classification model C is used for training discrimination parameters, and the trained discrimination model D is obtained through the trained discrimination parameters. Then, for a given discrimination model D, at this timeIs constant, only need to minimize/>And the prediction classification result of the classification model C is more accurate, so that the classification parameters are optimized and solved, the classification model C is updated, and the iteration is performed until the classification model C and the discrimination model D meet the preset conditions. The preset condition is equivalent to judging whether the classification model and the discrimination model reach preset requirements, the classification model is used for predicting whether a user has preference on a target object, and the discrimination model is used for judging whether an input sample is true or false, so that the preset condition can be set according to the accuracy of discrimination of the discrimination model in discrimination of the true or false, for example, when the accuracy of discrimination of the discrimination model in discrimination of the true or false reaches 0.5, the discrimination model and the classification model are considered to reach the preset condition, wherein if the accuracy of the discrimination model in discrimination of the true or false reaches 0.5, the discrimination model is difficult to discriminate the true or false, and the prediction accuracy of the classification model is very high.
In addition, the stochastic gradient algorithm may include a stochastic gradient ramp-up algorithm and a stochastic gradient ramp-down algorithm, in which, in minimizing the loss function, the minimized loss function and corresponding parameter values may be found by the gradient ramp-down concept, which in turn may be found by the gradient ramp-up concept if a maximized loss function is required.
In summary, the sample set corresponding to the discriminant model D is a true sample (x, y) and a counterfeit sampleThe user corresponding to the real sample (x, y) is selected from a marked sample set consisting of a seed user and a negative user, the real classification result corresponding to the seed user is y=1, the real classification result corresponding to the negative user is y=0, and the fake sample/>The corresponding user is selected from a classified sample set consisting of seed users and target users,/>Is a predictive classification result obtained by using the classification model C (x). The discrimination model D can discriminate the authenticity of the input sample. The classification model C (x) is used for learning how to identify the target user as a potentially extensible user, i.e. is used for predicting whether the user has a preference on the target object, so that the prediction result of the user is more accurate, and the discrimination model D is spoofed. In the embodiment of the invention, iterative training is continuously carried out on the classification model C (x) and the discrimination model D until the classification model C (x) and the discrimination model D reach counter balance, the discrimination model D is difficult to discriminate the authenticity of the input sample, and at the moment, the classification model C (x) can make predictions which accord with the distribution of the real sample for a target user.
In the embodiment of the invention, the specific user is subjected to semi-supervised training based on the antagonistic network learning algorithm, and the problem of how to train and classify by using a small number of marked samples and a large number of unmarked samples is mainly considered. It can be seen that the real sample corresponds to a marked sample, the counterfeit sample corresponds to an unmarked sample, the classification model C (x) corresponds to a user marked model, and the unmarked sample, i.e., the counterfeit sample, can be marked, so that semi-supervised training of a specific user can be realized, the adaptability threshold selection sensitivity problem of unsupervised learning and the over-fitting problem of supervised learning and a special optimization method are avoided, and the accuracy of the trained algorithm model is improved.
As a referenceable embodiment of the present invention, training the discrimination parameters by using a random gradient ascent method according to feature data of a specific user to obtain a discrimination model corresponding to the discrimination parameters may include:
step S1021, sampling at least one first classified sample set from a specific user;
Step S1022, inputting the feature data corresponding to at least one first classification sample set into a classification model to obtain a prediction classification result corresponding to at least one first classification sample set;
step S1023, sampling at least one marked sample set from the seed user and the negative user, and determining a real classification result corresponding to the at least one marked sample set;
step S1024, constructing a discrimination training set by using the predicted classification result corresponding to at least one first classification sample set and the real classification result corresponding to at least one marking sample set;
step S1025, training the discrimination parameters according to the discrimination training set by a random gradient ascent method.
In the embodiment of the invention, the seed user and the target user form a classification sample set, and in the method for training the discrimination model by using the classification model, the classification model is known, and the discrimination model is solved, namely the discrimination parameters of the discrimination model are solved. The stochastic gradient method is an optimization algorithm commonly used in machine learning and artificial intelligence to recursively approximate minimum deviation models and solve maxima or minima along the gradient ascending or descending direction. The random gradient method is to extract a batch of training data from a training set randomly to input and calculate the average gradient, and the method has the advantages of reducing the calculation cost of each iteration and accelerating the model training speed. As already described above, the discriminant model D can be obtained by maximum likelihood optimization V (D, C), that is, solving for maxima along the gradient-increasing direction.
Therefore, in the method for training discrimination parameters, a plurality of first classification sample sets (the first one is only for distinguishing from the second one hereinafter, without practical meaning) are sampled from the classification sample set composed of the seed user phi seed and the target user phi target, and the user characteristic data in the first classification sample set is input into the classification model C (x) to obtain the prediction classification resultAt least one marked sample set is sampled from the seed user phi seed and the negative user phi neg, and the true classification result corresponding to the user in the marked sample set is determined. Then, constructing a discrimination training set by utilizing the prediction classification result corresponding to the at least one first classification sample set and the real classification result corresponding to the at least one marking sample set:
Wherein Fake represents a counterfeit sample and Real represents a true sample. Finally, training a discrimination parameter theta D by a random gradient rising method, wherein a specific calculation formula is as follows:
In the embodiment of the invention, after the discrimination parameters are trained, the discrimination parameters can be fixed to obtain the corresponding discrimination model D, and then the classification parameters are updated by using the loss function of the discrimination model D, so that the aim of training the classification model C is fulfilled. The loss function has been described above as:
for a given discriminant model D, at this time Is constant and only needs to be minimizedThe prediction classification result of the classification model C can be more accurate, namely, the minimum value is solved along the gradient descending direction.
Thus, as a referenceable embodiment of the present invention, training classification parameters with a loss function of a discriminant model by a random gradient descent method may include: determining a loss function of the discrimination model; sampling at least one second set of classified samples from a particular user; and inputting the characteristic data corresponding to at least one second classification sample set into a loss function of the discrimination model, and training classification parameters by a random gradient descent method. In the method for training the classification parameters, a plurality of second classification sample sets (the second is only distinguished from the first classification sample set in the above description and has no practical meaning) are sampled from a classification sample set formed by a seed user phi seed and a target user phi target, and user characteristic data in the second classification sample set is input into a classification model C (x) to obtain a prediction classification resultThen training the classification parameter theta C by a random gradient descent method, wherein the specific calculation formula is as follows:
FIG. 4 is a schematic diagram of the main flow of a method of training a classification model and a discriminant model according to an embodiment of the present invention. As shown in fig. 4, the main flow of the method of training the classification model and the discriminant model may include:
Step S401, initializing a classification model of an countermeasure network learning algorithm;
step S402, sampling at least one first classified sample set from a specific user;
step S403, inputting the feature data corresponding to at least one first classification sample set into a classification model to obtain a prediction classification result corresponding to at least one first classification sample set;
Step S404, sampling at least one marked sample set from the seed user and the negative user, and determining a real classification result corresponding to the at least one marked sample set;
Step S405, constructing a discrimination training set by using a prediction classification result corresponding to at least one first classification sample set and a real classification result corresponding to at least one marking sample set;
Step S406, training the discrimination parameters according to the discrimination training set by a random gradient ascending method to obtain a discrimination model and a loss function corresponding to the discrimination parameters;
step S407, sampling at least one second classification sample set from the specific user;
step S408, inputting the feature data corresponding to at least one second classification sample set into a loss function of a discrimination model, training classification parameters by a random gradient descent method, and updating the classification model according to the trained classification parameters;
Step S409, judging whether the updated classification model and the trained discrimination model meet preset conditions, if yes, executing step S410, and if not, executing step S402 to step S408 again for iterative training;
In step S410, a target classification model and a target discrimination model of the challenge network learning algorithm are obtained.
In the embodiment of the invention, the classification model and the discrimination model are trained by using the countermeasure network learning algorithm, so that all users can be supervised and trained, and the problems of sensitivity of adaptability threshold selection of unsupervised learning, supervised learning and over-fitting of special optimization methods are solved. The classification model is used for classifying the target user and outputting the predicted probability distribution of the target user, which is equivalent to a generator of the countermeasure network, and the role of the discrimination model is to discriminate whether the predicted probability distribution of the classification model is identical with the real probability distribution, so that the classification model and the countermeasure balance of the discrimination model enable the classification model to learn the joint distribution from the seed user and the target user. In addition, the iterative training is carried out by adopting a random gradient algorithm, so that the calculation cost of each iteration can be reduced, and the model training speed is increased.
It should be noted that in step S401, the classification model is initialized first, then the discrimination model is trained by using the initialized classification model, then the classification model is trained by using the obtained discrimination model again, and iterative training is continuously performed. Therefore, it is important to initialize the acquisition of the classification model, and as a referenceable embodiment of the present invention, initializing the classification model against the network learning algorithm may include: and obtaining an initial value of the classification parameter, and directly initializing by using the initial value of the classification parameter. That is, the initial value of the classification parameter is directly given, and the initial value is substituted into the classification model to complete the initialization process of the classification model.
In order to maintain the overall stability of the training process, the classification model can be trained in advance to maintain a certain accuracy, and then the training system is connected to the countermeasure network for overall training. The method for pre-training the classification model can be as follows: at least one classification training set is sampled from the seed user and the negative user, and at least one initial classification model is trained in advance by using the at least one classification training set so as to finish initialization of the classification model. Considering that the seed user and the negative user belong to marked users, the real classification result corresponding to the seed user is y=1, and the real classification result corresponding to the negative user is y=0, so in the embodiment of the invention, the seed user and the negative user can be utilized to construct a classification training set, and the constructed classification training set is utilized to train an initial classification model in advance.
In addition, in the embodiment of the invention, at least one classification training set can be constructed. Then, for each classification training set, pre-training an initial classification model corresponding to the classification training set, and further obtaining a corresponding classification model and a corresponding discrimination model. Therefore, as a referenceable embodiment of the present invention, obtaining a target classification model and a target discrimination model against a network learning algorithm may include: obtaining at least one optional classification model and at least one optional discrimination model corresponding to the at least one initial classification model; from the at least one selectable classification model and the at least one selectable discriminant model, a target classification model and a target discriminant model are determined. In general, different classification training sets can train different classification models, vote on prediction classification results corresponding to different classification models, and finally select a target classification model.
Fig. 5 is a schematic structural diagram of a similar crowd expansion system according to an embodiment of the invention, as shown in fig. 5, the structure of the similar crowd expansion system according to the embodiment of the invention may include: a data processing section, a classification model section, and a discrimination model section.
The data processing section may generate the feature data of the user based on the basic attribute data of the user and the behavior data of the user. The basic attribute data of the user represent long-term characteristics of the user, the behavior data of the user represent recent characteristics of the user, and the behavior data of the user are sparse, so that the user needs to be processed. The behavior data of the user is divided into a large number of common click behavior data and a relatively small number of other behavior data, such as attention behavior, collection behavior, purchasing behavior and purchasing behavior. In the embodiment of the present invention, the feature embedding process may be performed on the ordinary click behavior data, which has been described in detail in the above steps S10121 to S10124, and will not be described here again; the word segmentation feature processing may be performed on other behavior data of relatively small numbers, which have been described in detail in the above steps S10131 to S10313, and will not be described here.
The classification model portion and the discrimination model portion constitute a countermeasure system, and countermeasure balance of both enables the classification model to learn joint distribution from the seed user and the target user. The sample set corresponding to the discrimination model is a true sample (x, y) and a fake sampleThe user corresponding to the real sample (x, y) is selected from a marked sample set consisting of a seed user and a negative user, the real classification result corresponding to the seed user is y=1, the real classification result corresponding to the negative user is y=0, and the fake sample/>The corresponding user is selected from a classified sample set consisting of seed users and target users,/>The prediction classification result is obtained by using the classification model. The discrimination model can discriminate the authenticity of the input sample. The classification model is used for learning how to identify the target user as a potential extensible user, namely, the classification model is used for predicting whether the user has preference on the target object, so that the prediction result of the classification model is more accurate, and the discrimination model is cheated. And (3) continuously performing iterative training on the classification model and the discrimination model until the classification model and the discrimination model reach countermeasure balance, wherein the discrimination model D is difficult to discriminate the authenticity of the input sample, and the classification model can make predictions conforming to the distribution of the real sample for the target user. The process of training the classification model and the discrimination model has been described in detail in the above steps S401 to S410, and will not be described here.
Fig. 6 is a schematic diagram of the main modules of a similar crowd expansion device according to an embodiment of the invention. As shown in fig. 6, the similar crowd expanding device 600 of the embodiment of the invention mainly comprises the following modules: a generation module 601, a training module 602, and a determination module 603.
The generating module 601 may be configured to obtain a specific user corresponding to the target object, generate feature data of the specific user according to basic attribute data and behavior data of the specific user, where the specific user may include: seed users and target users; the training module 602 may be configured to train an algorithm model of an antagonistic network learning algorithm using a stochastic gradient method according to the feature data of the specific user; the determining module 603 may be configured to determine, based on the algorithm model, the extensible user corresponding to the target object from the target users according to the feature data of the target users.
In an embodiment of the present invention, training module 602 may also be configured to: initializing a classification model of an countermeasure network learning algorithm; training the discrimination parameters according to the characteristic data and the classification model of the specific user by a random gradient ascending method to obtain a discrimination model corresponding to the discrimination parameters; training classification parameters by using a loss function of the discrimination model through a random gradient descent method, and updating the classification model according to the trained classification parameters; judging whether the judging model and the updated classifying model accord with preset conditions or not; if yes, determining the updated classification model as a target classification model, and determining the discrimination model as a target discrimination model; if not, carrying out model training by utilizing the characteristic data of the specific user and the updated classification model until the discrimination model and the classification model obtained by training meet the preset conditions.
In an embodiment of the present invention, training module 602 may also be configured to: sampling at least one first set of classified samples from a particular user; inputting the characteristic data corresponding to the at least one first classification sample set into a classification model to obtain a prediction classification result corresponding to the at least one first classification sample set; sampling at least one marked sample set from a seed user and a negative user, and determining a real classification result corresponding to the at least one marked sample set, wherein the real classification result corresponding to the seed user is 1, and the real classification result corresponding to the negative user is 0; constructing a discrimination training set by utilizing a prediction classification result corresponding to at least one first classification sample set and a real classification result corresponding to at least one marking sample set; and training the discrimination parameters according to the discrimination training set by a random gradient rising method.
In an embodiment of the present invention, training module 602 may also be configured to: determining a loss function of the discrimination model; sampling at least one second set of classified samples from a particular user; and inputting the characteristic data corresponding to at least one second classification sample set into a loss function of the discrimination model, and training classification parameters by a random gradient descent method.
In an embodiment of the present invention, training module 602 may also be configured to: acquiring an initial value of the classification parameter, and initializing by directly using the initial value of the classification parameter; and sampling at least one classification training set from the seed user and the negative user, and pre-training at least one initial classification model by using the at least one classification training set to finish initialization of the classification model.
In an embodiment of the present invention, training module 602 may also be configured to: obtaining at least one optional classification model and at least one optional discrimination model corresponding to the at least one initial classification model; from the at least one selectable classification model and the at least one selectable discriminant model, a target classification model and a target discriminant model are determined.
In the embodiment of the present invention, the similar crowd expanding device may further include: a sampling module (not shown). The sampling module may be used to: randomly sampling negative users from target users; sampling the negative user from the target user according to the seed user behavior data and the second behavior data of the seed user; the first behavior data of the negative user and the first behavior data of the seed user are intersected, and the second behavior data of the negative user and the second behavior data of the seed user are not intersected.
In the embodiment of the present invention, the behavior data may include: first behavior data and second behavior data; the generating module 601 may be further configured to: acquiring basic attribute data, first behavior data and second behavior data; processing the first line of data based on a preset embedded feature processing rule to generate embedded feature data corresponding to the first line of data; processing the second behavior data based on a preset word segmentation feature processing rule to generate word segmentation feature data corresponding to the second behavior data; and combining the basic attribute data, the embedded feature data and the word segmentation feature data to generate feature data.
In the embodiment of the present invention, the generating module 601 may further be configured to: acquiring at least one item attribute data corresponding to the first row of data; according to a preset time threshold corresponding to the at least one item attribute data, carrying out sectional processing on the at least one item attribute data to obtain a behavior sequence corresponding to the at least one item attribute data; embedding the behavior sequence corresponding to the at least one item attribute data by utilizing a word vector embedding algorithm to obtain sub-embedded feature data corresponding to the at least one item attribute data; and combining the sub-embedded feature data corresponding to the at least one item attribute data to generate embedded feature data corresponding to the first row of data.
In the embodiment of the present invention, the generating module 601 may further be configured to: acquiring an article description sentence corresponding to the second behavior data; performing word segmentation processing on the object description sentence to obtain at least one word segment, and filtering and screening the at least one word segment; and generating word segmentation characteristic data corresponding to the second behavior data by utilizing the filtered word segmentation.
In the embodiment of the present invention, the determining module 603 may further be configured to: inputting the characteristic data of the target user into a target classification model to obtain a prediction classification result corresponding to the target user; based on preset expansion conditions, expandable users are selected from target users according to prediction classification results corresponding to the target users.
From the above description, it can be seen that the similar crowd expansion device of the embodiment of the invention can generate feature data by using the basic attribute data and behavior data of a specific user, and has important significance for improving the conversion rate due to the introduction of the behavior data, so that the accuracy of a trained algorithm model can be improved; then training an algorithm model of an anti-network learning algorithm by combining characteristic data of a specific user through a random gradient method, and performing semi-supervised training on the specific user based on the anti-network learning algorithm, wherein the method is different from a method for performing supervised training by only using seed users and a method based on unsupervised clustering by all users, so that the adaptive threshold selection sensitivity problem of unsupervised learning and the overfitting problem of supervised learning and a special optimization method are avoided; and finally, selecting an extensible user from target users by using a trained algorithm model, and completing similar crowd extension.
In addition, the generation module of the similar crowd expansion device of the embodiment of the invention can comprehensively consider the long-term characteristic and the recent behavior characteristic of the user from three aspects of basic attribute data, first behavior data and second behavior data, so that the accuracy of the characteristic data and an algorithm model can be improved, the accuracy of the obtained expandable user can be further ensured, and in addition, the introduction of the behavior characteristic of the user has important significance for improving the conversion rate. And taking the characteristics of the first behavior data, such as a large quantity and sparse characteristics into consideration, adopting an embedded characteristic processing method to obtain corresponding embedded characteristic data, and avoiding the problem of overfitting of a trained model.
In addition, the training module of the similar crowd expansion device provided by the embodiment of the invention can train the classification model and the discrimination model by utilizing the antagonistic network learning algorithm, and supervise and train all users, so that the problems of sensitivity of the adaptive threshold selection of unsupervised learning, supervised learning and over-fitting of a special optimization method are solved. The classification model is used for classifying the target user and outputting the predicted probability distribution of the target user, which is equivalent to a generator of the countermeasure network, and the role of the discrimination model is to discriminate whether the predicted probability distribution of the classification model is identical with the real probability distribution, so that the classification model and the countermeasure balance of the discrimination model enable the classification model to learn the joint distribution from the seed user and the target user. In addition, the iterative training is carried out by adopting a random gradient algorithm, so that the calculation cost of each iteration can be reduced, and the model training speed is increased.
Fig. 7 illustrates an exemplary system architecture 700 to which the similar crowd expansion method or similar crowd expansion means of embodiments of the invention may be applied.
As shown in fig. 7, a system architecture 700 may include terminal devices 701, 702, 703, a network 704, and a server 705. The network 704 is the medium used to provide communication links between the terminal devices 701, 702, 703 and the server 705. The network 704 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.
A user may interact with the server 705 via the network 704 using the terminal devices 701, 702, 703 to receive or send messages or the like. Various communication client applications such as shopping class applications, web browser applications, search class applications, instant messaging tools, mailbox clients, social platform software, etc. (by way of example only) may be installed on the terminal devices 701, 702, 703.
The terminal devices 701, 702, 703 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smartphones, tablets, laptop and desktop computers, and the like.
The server 705 may be a server providing various services, such as a background management server (by way of example only) providing support for shopping-type websites browsed by users using the terminal devices 701, 702, 703. The background management server may analyze and process the received data such as the product information query request, and feedback the processing result (e.g., the target push information, the product information—only an example) to the terminal device.
It should be noted that, the similar crowd expansion method provided in the embodiment of the present invention is generally executed by the server 705, and accordingly, the similar crowd expansion device is generally disposed in the server 705.
It should be understood that the number of terminal devices, networks and servers in fig. 7 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.
Referring now to FIG. 8, there is illustrated a schematic diagram of a computer system 800 suitable for use in implementing an embodiment of the present invention. The terminal device shown in fig. 8 is only an example, and should not impose any limitation on the functions and the scope of use of the embodiment of the present invention.
As shown in fig. 8, the computer system 800 includes a Central Processing Unit (CPU) 801 that can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) 802 or a program loaded from a storage section 808 into a Random Access Memory (RAM) 803. In the RAM803, various programs and data required for the operation of the system 800 are also stored. The CPU 801, ROM 802, and RAM803 are connected to each other by a bus 804. An input/output (I/O) interface 805 is also connected to the bus 804.
The following components are connected to the I/O interface 805: an input portion 806 including a keyboard, mouse, etc.; an output portion 807 including a display such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and a speaker; a storage section 808 including a hard disk or the like; and a communication section 809 including a network interface card such as a LAN card, a modem, or the like. The communication section 809 performs communication processing via a network such as the internet. The drive 810 is also connected to the I/O interface 805 as needed. A removable medium 811 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 810 as needed so that a computer program read out therefrom is mounted into the storage section 808 as needed.
In particular, according to embodiments of the present disclosure, the processes described above with reference to flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method shown in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network via the communication section 809, and/or installed from the removable media 811. The above-described functions defined in the system of the present invention are performed when the computer program is executed by a Central Processing Unit (CPU) 801.
The computer readable medium shown in the present invention may be a computer readable signal medium or a computer readable storage medium, or any combination of the two. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present invention, however, the computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, with the computer-readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, RF, etc., or any suitable combination of the foregoing.
The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
The modules involved in the embodiments of the present invention may be implemented in software or in hardware. The described modules may also be provided in a processor, for example, as: a processor includes a generation module, a training module, and a determination module. The names of the modules do not limit the module itself in some cases, for example, the generation module may also be described as "a module for acquiring a specific user corresponding to the target object, and generating feature data of the specific user according to basic attribute data and behavior data of the specific user".
As another aspect, the present invention also provides a computer-readable medium that may be contained in the apparatus described in the above embodiments; or may be present alone without being fitted into the device. The computer readable medium carries one or more programs which, when executed by a device, cause the device to include: acquiring a specific user corresponding to the target object, generating feature data of the specific user according to basic attribute data and behavior data of the specific user, wherein the specific user can comprise: seed users and target users; training an algorithm model of an antagonistic network learning algorithm by adopting a random gradient method according to the characteristic data of a specific user; and determining the extensible user corresponding to the target object from the target user based on the algorithm model according to the characteristic data of the target user.
According to the technical scheme provided by the embodiment of the invention, the characteristic data can be generated by utilizing the basic attribute data and the behavior data of the specific user, and the behavior data is introduced, so that the method has important significance for improving the conversion rate, and the accuracy of a trained algorithm model can be improved; then training an algorithm model of an anti-network learning algorithm by combining characteristic data of a specific user through a random gradient method, and performing semi-supervised training on the specific user based on the anti-network learning algorithm, wherein the method is different from a method for performing supervised training by only using seed users and a method based on unsupervised clustering by all users, so that the adaptive threshold selection sensitivity problem of unsupervised learning and the overfitting problem of supervised learning and a special optimization method are avoided; and finally, selecting an extensible user from target users by using a trained algorithm model, and completing similar crowd extension.
In addition, in the method for generating the feature data, the long-term feature and the recent feature of the user can be comprehensively considered from three aspects of basic attribute data, first behavior data and second behavior data, so that the accuracy of the feature data and an algorithm model can be improved, the accuracy of the obtained expandable user can be further ensured, and in addition, the introduction of the user has important significance for improving the conversion rate by taking the feature of the user into consideration. And taking the characteristics of the first behavior data, such as a large quantity and sparse characteristics into consideration, adopting an embedded characteristic processing method to obtain corresponding embedded characteristic data, and avoiding the problem of overfitting of a trained model.
In addition, in the method for training the classification model and the discrimination model, which is disclosed by the embodiment of the invention, the classification model and the discrimination model can be trained by using the countermeasure network learning algorithm, and all users can be supervised and trained, so that the problems of sensitivity of adaptability threshold selection of unsupervised learning and overfitting of supervised learning and special optimization methods are solved. The classification model is used for classifying the target user and outputting the predicted probability distribution of the target user, which is equivalent to a generator of the countermeasure network, and the role of the discrimination model is to discriminate whether the predicted probability distribution of the classification model is identical with the real probability distribution, so that the classification model and the countermeasure balance of the discrimination model enable the classification model to learn the joint distribution from the seed user and the target user. In addition, the iterative training is carried out by adopting a random gradient algorithm, so that the calculation cost of each iteration can be reduced, and the model training speed is increased.
The above embodiments do not limit the scope of the present invention. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives can occur depending upon design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of the present invention.

Claims (13)

1. A method for expanding a similar crowd, comprising:
acquiring a specific user corresponding to a target object, and generating characteristic data of the specific user according to basic attribute data and behavior data of the specific user, wherein the specific user comprises: seed users and target users;
Training an algorithm model of an antagonistic network learning algorithm by adopting a random gradient method according to the characteristic data of the specific user;
based on the algorithm model, determining an extensible user corresponding to the target object from the target users according to the characteristic data of the target users;
The training the algorithm model of the countermeasure network learning algorithm by adopting a random gradient method according to the characteristic data of the specific user comprises the following steps:
initializing a classification model of an countermeasure network learning algorithm;
Training a discrimination parameter according to the characteristic data of the specific user and the classification model by a random gradient ascending method to obtain a discrimination model corresponding to the discrimination parameter;
training classification parameters by using a random gradient descent method and utilizing a loss function of the discrimination model, and updating the classification model according to the trained classification parameters;
Judging whether the judging model and the updated classifying model accord with preset conditions or not;
If yes, determining the updated classification model as a target classification model, and determining the discrimination model as a target discrimination model;
If not, carrying out model training by utilizing the characteristic data of the specific user and the updated classification model until the discrimination model and the classification model obtained by training meet preset conditions.
2. The method according to claim 1, wherein training a discrimination parameter according to the feature data of the specific user and the classification model by a random gradient ascent method, and obtaining a discrimination model corresponding to the discrimination parameter, comprises:
Sampling at least one first set of classified samples from the particular user;
inputting the feature data corresponding to the at least one first classification sample set into the classification model to obtain a prediction classification result corresponding to the at least one first classification sample set;
Sampling at least one marked sample set from the seed user and the negative user, and determining a real classification result corresponding to the at least one marked sample set, wherein the real classification result corresponding to the seed user is 1, and the real classification result corresponding to the negative user is 0;
Constructing a discrimination training set by utilizing a prediction classification result corresponding to the at least one first classification sample set and a real classification result corresponding to the at least one mark sample set;
Training the discrimination parameters according to the discrimination training set by a random gradient rising method.
3. The method according to claim 1, wherein training classification parameters using a loss function of the discriminant model by a random gradient descent method comprises:
determining a loss function of the discrimination model;
sampling at least one second set of classified samples from the particular user;
And inputting the characteristic data corresponding to the at least one second classification sample set into a loss function of the discrimination model, and training the classification parameters by a random gradient descent method.
4. The method of claim 1, wherein initializing the classification model against the network learning algorithm comprises:
Acquiring an initial value of a classification parameter, and initializing by directly utilizing the initial value of the classification parameter; and
At least one classification training set is sampled from the seed user and the negative user, and at least one initial classification model is trained in advance by utilizing the at least one classification training set so as to finish initialization of the classification model.
5. The method of claim 4, wherein the obtaining the object classification model and the object discrimination model for the challenge network learning algorithm comprises:
Obtaining at least one optional classification model and at least one optional discriminant model corresponding to the at least one initial classification model;
Determining the target classification model and the target discrimination model from the at least one selectable classification model and the at least one selectable discrimination model.
6. The method according to claim 2 or 4, characterized in that the method further comprises:
Randomly sampling the negative user from the target users; and
Sampling the negative user from the target user according to the seed user behavior data and the second behavior data of the seed user; wherein,
The first behavior data of the negative user and the first behavior data of the seed user are intersected, and the second behavior data of the negative user and the second behavior data of the seed user are not intersected.
7. The method of claim 1, wherein the behavioral data comprises: first behavior data and second behavior data; and
The generating the characteristic data of the specific user according to the basic attribute data and the behavior data of the specific user comprises the following steps:
acquiring the basic attribute data, the first behavior data and the second behavior data;
Processing the first line of data based on a preset embedded feature processing rule to generate embedded feature data corresponding to the first line of data;
Processing the second behavior data based on a preset word segmentation feature processing rule to generate word segmentation feature data corresponding to the second behavior data;
and combining the basic attribute data, the embedded feature data and the word segmentation feature data to generate the feature data.
8. The method of claim 7, wherein the processing the first row of data based on the preset embedded feature processing rule to generate the embedded feature data corresponding to the first row of data comprises:
acquiring at least one item attribute data corresponding to the first row of data;
according to a preset time threshold corresponding to the at least one item attribute data, carrying out segmentation processing on the at least one item attribute data to obtain a behavior sequence corresponding to the at least one item attribute data;
Embedding the behavior sequence corresponding to the at least one item attribute data by utilizing a word vector embedding algorithm to obtain sub-embedded feature data corresponding to the at least one item attribute data;
and combining the sub-embedded feature data corresponding to the at least one item attribute data to generate the embedded feature data corresponding to the first row of data.
9. The method of claim 7, wherein the processing the second behavior data based on the preset word segmentation feature processing rule to generate word segmentation feature data corresponding to the second behavior data includes:
Acquiring an article description sentence corresponding to the second behavior data;
performing word segmentation processing on the article description sentence to obtain at least one word segment, and filtering and screening the at least one word segment;
and generating word segmentation characteristic data corresponding to the second behavior data by utilizing the filtered word segmentation.
10. The method according to claim 1, wherein the determining, based on the algorithm model, the extensible user corresponding to the target object from the target users according to the feature data of the target users includes:
inputting the characteristic data of the target user into the target classification model to obtain a prediction classification result corresponding to the target user;
and selecting the expandable user from the target users according to the prediction classification result corresponding to the target users based on preset expansion conditions.
11. A similar crowd expansion device, comprising:
The generating module is used for acquiring a specific user corresponding to the target object, generating characteristic data of the specific user according to basic attribute data and behavior data of the specific user, wherein the specific user comprises: seed users and target users;
The training module is used for training an algorithm model of an antagonistic network learning algorithm by adopting a random gradient method according to the characteristic data of the specific user; the method is particularly used for: initializing a classification model of an countermeasure network learning algorithm; training a discrimination parameter according to the characteristic data of the specific user and the classification model by a random gradient ascending method to obtain a discrimination model corresponding to the discrimination parameter; training classification parameters by using a random gradient descent method and utilizing a loss function of the discrimination model, and updating the classification model according to the trained classification parameters; judging whether the judging model and the updated classifying model accord with preset conditions or not; if yes, determining the updated classification model as a target classification model, and determining the discrimination model as a target discrimination model; if not, carrying out model training by utilizing the characteristic data of the specific user and the updated classification model until the discrimination model and the classification model obtained by training meet preset conditions;
and the determining module is used for determining the extensible user corresponding to the target object from the target users according to the characteristic data of the target users based on the algorithm model.
12. An electronic device, comprising:
One or more processors;
Storage means for storing one or more programs,
When executed by the one or more processors, causes the one or more processors to implement the method of any of claims 1-10.
13. A computer readable medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method according to any of claims 1-10.
CN202010580213.1A 2020-06-23 2020-06-23 Similar crowd expansion method and device Active CN113762298B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010580213.1A CN113762298B (en) 2020-06-23 2020-06-23 Similar crowd expansion method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010580213.1A CN113762298B (en) 2020-06-23 2020-06-23 Similar crowd expansion method and device

Publications (2)

Publication Number Publication Date
CN113762298A CN113762298A (en) 2021-12-07
CN113762298B true CN113762298B (en) 2024-06-18

Family

ID=78785422

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010580213.1A Active CN113762298B (en) 2020-06-23 2020-06-23 Similar crowd expansion method and device

Country Status (1)

Country Link
CN (1) CN113762298B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114792256B (en) * 2022-06-23 2023-05-26 上海维智卓新信息科技有限公司 Crowd expansion method and device based on model selection

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110019770A (en) * 2017-07-24 2019-07-16 华为技术有限公司 The method and apparatus of train classification models
CN110532377B (en) * 2019-05-13 2021-09-14 南京大学 Semi-supervised text classification method based on confrontation training and confrontation learning network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Similar Handwritten Chinese Character Recognition Using Hierarchical CNN Model;Qingqing Wang;IEEE;20180129;全文 *
结合半监督学习和LDA模型的文本分类方法;韩栋;王春华;肖敏;;计算机工程与设计;20181016(第10期);全文 *

Also Published As

Publication number Publication date
CN113762298A (en) 2021-12-07

Similar Documents

Publication Publication Date Title
CN109063163B (en) Music recommendation method, device, terminal equipment and medium
US11568315B2 (en) Systems and methods for learning user representations for open vocabulary data sets
CN110069709B (en) Intention recognition method, device, computer readable medium and electronic equipment
CN108108743B (en) Abnormal user identification method and device for identifying abnormal user
CN110837862B (en) User classification method and device
CN111667022A (en) User data processing method and device, computer equipment and storage medium
CN112395487B (en) Information recommendation method and device, computer readable storage medium and electronic equipment
CN109471978B (en) Electronic resource recommendation method and device
CN105225135B (en) Potential customer identification method and device
CN113190670A (en) Information display method and system based on big data platform
CN114298122B (en) Data classification method, apparatus, device, storage medium and computer program product
CN111191825A (en) User default prediction method and device and electronic equipment
CN112836509A (en) Expert system knowledge base construction method and system
CN112418320A (en) Enterprise association relation identification method and device and storage medium
Aziguli et al. A robust text classifier based on denoising deep neural network in the analysis of big data
CN111209469A (en) Personalized recommendation method and device, computer equipment and storage medium
Lee et al. Effective evolutionary multilabel feature selection under a budget constraint
WO2023024408A1 (en) Method for determining feature vector of user, and related device and medium
CN113392920B (en) Method, apparatus, device, medium, and program product for generating cheating prediction model
CN113051911B (en) Method, apparatus, device, medium and program product for extracting sensitive words
CN113762298B (en) Similar crowd expansion method and device
Mathivanan et al. A comparative study on dimensionality reduction between principal component analysis and k-means clustering
JPWO2020095357A1 (en) Search needs evaluation device, search needs evaluation system, and search needs evaluation method
US20230229691A1 (en) Methods and systems for prediction of a description and icon for a given field name
CN113822390B (en) User portrait construction method and device, electronic equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant