User classification method, device, electronic device and storage medium

Abstract

The present application discloses a user classification method, device, electronic device and storage medium, the method comprising: obtaining the word class features of each word class based on user text data, wherein the word class is obtained based on keyword clustering in the user text data; generating a word class feature vector for each user based on the generated word class features and the relationship between each user and the word class, wherein the word class feature vector is an N-dimensional vector, where N is the total number of word classes; obtaining the probability distribution of each user with respect to preset topics based on the word class feature vector of each user; for each user, generating the probability distribution of the user with respect to preset business scenarios based on the mapping relationship between the preset word class and the preset business scenarios, the probability distribution of the word class under the preset topics, and the probability distribution of the user with respect to the preset topics.

Description

Technical Field

[0001] The present application belongs to the field of computer technology, and specifically relates to a user classification method, device, electronic device and storage medium. Background Art

[0002] With the gradual increase in the number of Internet users, refined operations have become the mainstream operation method. How to finely classify users and use segmented operation strategies to accurately push suitable products to users has become an urgent problem to be solved by every product or operation personnel.

[0003] In the prior art, existing portrait tags or single keywords are mainly used as input, and the k-means method is used to classify users.

[0004] However, since some scenarios have a rich variety of vocabulary, while others have fewer, or there are many synonyms, similar words, or new words, using keywords as direct input will result in greater instability or potential bias in the final output. For example, using keywords in a hot scenario or popular keywords as input will result in the final output tending to favor the information expressed by these keywords, and the output may be biased, leading to inaccurate classification results. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a user classification method, device, electronic device and storage medium, which can solve the problem of inaccurate classification results in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a user classification method, the method comprising:

[0007] According to the user text data, obtaining a word class feature of each word class, wherein the word class is obtained based on keyword clustering in the user text data;

[0008] Generate a word class feature vector for each user based on the generated word class features and the relationship between each user and the word class, wherein the word class feature vector is an N-dimensional vector, where N is the total number of word classes;

[0009] According to each user's word class feature vector, obtain each user's probability distribution on each preset topic;

[0010] For each user, the probability distribution of the user regarding the preset business scenarios is generated based on the mapping relationship between the preset word classes and the preset business scenarios, the probability distribution of the word classes under the preset topics, and the probability distribution of the user regarding the preset topics.

[0011] In a second aspect, an embodiment of the present application provides a user classification device, the device comprising:

[0012] A first acquisition module, configured to acquire a word class feature of each word class according to user text data, wherein the word class is obtained by clustering keywords in the user text data;

[0013] A first generating module, configured to generate a word class feature vector for each user based on the generated word class features and the relationship between each user and the word class, wherein the word class feature vector is an N-dimensional vector, where N is the total number of word classes;

[0014] The second acquisition module is used to obtain the probability distribution of each user on each preset topic according to the word class feature vector of each user;

[0015] The second generation module is used to generate, for each user, the probability distribution of the user regarding the preset business scenarios based on the mapping relationship between the preset word classes and the preset business scenarios, the probability distribution of the word classes under the preset topics, and the probability distribution of the user regarding the preset topics.

[0016] In a third aspect, an embodiment of the present application provides an electronic device, which includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method described in the first aspect.

[0017] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0018] In a fifth aspect, an embodiment of the present application provides a chip, comprising a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run a program or instruction to implement the method described in the first aspect.

[0019] In the embodiment of the present application, for the user group to be classified, keywords in the user text data generated by the user group are extracted, the keywords are clustered to obtain multiple word classes, and the word class features of each word class are generated. Then, a word class feature vector is generated for each user, and the probability distribution of each user on each topic is obtained based on the word class feature vector of each user; finally, based on the mapping relationship between the topic, word class and business scenario, and the probability distribution of each user on each topic, the probability distribution of each user in the user group on each business scenario is generated.

[0020] Compared with the prior art, in the embodiment of the present application, keywords are classified, and word classes are used instead of single words as input features to classify user groups, which can reduce the head effect of hot scenes or popular words, improve the stability of input and the contribution of long-tail information dimensions, thereby improving the accuracy of user classification. In addition, using word classes as feature input units, establishing a low-dimensional mapping of topics, word classes and business scenarios, and classifying user groups can be achieved. When finally recognizing users, it can be switched from not only the user perspective but also the business scenario perspective, which can be efficient, stable and converge to the business direction to automatically describe users, and quickly help product operations recognize users. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a flow chart of a user classification method provided in an embodiment of the present application;

[0022] Figure 2 This is a first example diagram of the calculation process of step 104 provided in the embodiment of the present application;

[0023] Figure 3 is a second example diagram of the calculation process of step 104 provided in an embodiment of the present application;

[0024] Figure 4 is a third example diagram of the calculation process of step 104 provided in an embodiment of the present application;

[0025] Figure 5 is a flow chart of another user classification method provided in an embodiment of the present application;

[0026] Figure 6 This is an example diagram of a word cloud diagram provided in an embodiment of the present application;

[0027] Figure 7 is a flow chart of another user classification method provided in an embodiment of the present application;

[0028] Figure 8 is a flow chart of another user classification method provided in an embodiment of the present application;

[0029] Fig. 9 is a structural block diagram of a user classification device provided in an embodiment of the present application;

[0030] Fig.10 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application;

[0031] Fig.11 It is a schematic diagram of the hardware structure of an electronic device implementing various embodiments of the present application. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0033] The terms "first", "second", etc. in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0034] With the gradual increase in Internet users, refined operations have become a very popular concept. How to quickly classify users and formulate corresponding business strategies is a problem that every product or operation often faces. In most scenarios, the common business method is to define rules for user classification based on their own experience, and then conduct user recognition; the second common method is to simply use existing tags or single keywords as input, use the k-means method to classify users, and then form new tags for recognition.

[0035] The division of populations based on manual business experience takes a relatively long time, and the output user cognition is limited by the rules of business experience output, and it is impossible to find new populations heuristically. Using portrait tags or keywords as direct input can solve the limitations of manual experience, but there are also the following problems:

[0036] 1) The population output is not fixed because the initial clustering point is randomly selected and k-means uses Euclidean distance to measure population similarity, which is easily affected by abnormal points; 2) Keywords as direct input will have greater instability or potential tendency because some scenarios have rich vocabulary types, while others have fewer, or there are more synonyms / similar words / new words. In this way, the final output will tend to be biased towards the information expressed by these keywords, and the output may be subjective; 3) The output population may not correspond to the existing business system, and there is a lack of a business scenario mapping process.

[0037] In order to solve the above technical problems, the embodiments of the present application provide a user classification method, device, electronic device and storage medium.

[0038] The user classification method provided in the embodiment of the present application is described in detail below through specific embodiments and application scenarios in conjunction with the accompanying drawings.

[0039] To facilitate understanding, some concepts involved in the embodiments of the present application are first described.

[0040] LDA (Latent Dirichlet allocation) is an algorithm model that calculates probabilities based on text topics and outputs the probability distribution of each topic.

[0041] PCA (Principal component analysis) uses orthogonal transformation to linearly transform the observed values ​​of a series of possibly correlated variables, thereby projecting them into a series of linearly uncorrelated variable values, which are the main components.

[0042] Information contribution refers to information gain, which is the degree to which information complexity (or uncertainty) is reduced under a certain condition.

[0043] TGI (Target Group Index) is an index that reflects the strength or weakness of the target group within a specific research scope. It is a statistical indicator of relative quantitative description.

[0044] An entity refers to the integration of ontology, instance and relationship in the graph, that is, the inherent existence, such as people, things and objects, while an instance refers to a specific person or thing, and a relationship refers to all the attributes of the entity, such as the portrait attribute of a person.

[0045] The term "scenario" originally refers to certain conditions when demand arises, including but not limited to time, space, equipment environment, etc. In the embodiments of the present application, these factors are integrated into demand types, such as meal delivery, enlightenment education, live broadcast of events, etc.

[0046] Next, the user classification method provided in the embodiment of the present application is introduced.

[0047] Figure 1 is a flow chart of a user classification method provided in an embodiment of the present application, such as Figure 1 As shown, the method may include the following steps: step 101, step 102, step 103 and step 104, wherein:

[0048] In step 101, a word class feature of each word class is obtained based on user text data, wherein the word class is obtained based on keyword clustering in the user text data.

[0049] In an embodiment of the present application, for a user group to be classified, user text data generated by the user behavior of each user in the user group is first obtained, and then keywords in the user text data are extracted, the extracted keywords are clustered to obtain multiple word classes, and the word class features of each word class are obtained.

[0050] In actual applications, user behaviors may include: opening the APP, searching in the APP, shopping in the APP, etc.

[0051] In an embodiment of the present application, when extracting keywords from user text data, it can be achieved through the following two steps: 1) performing data cleaning and word segmentation processing on the user text data in sequence to obtain multiple words; 2) filtering the multiple words according to preset word filtering rules to obtain multiple keywords, wherein the preset word filtering rules include: removing words without substantive meaning, removing cold words and hot words, and removing words with small weights.

[0052] In the embodiment of the present application, by performing data cleaning on the user text data, identifiable errors in the user text data can be corrected, and then the cleaned user text data is segmented to divide long sentences into multiple short words for subsequent extraction of valid keywords. In practical applications, data cleaning may include: checking data consistency, processing invalid values ​​and missing values, etc.

[0053] In the embodiment of the present application, for the multiple words obtained by the word segmentation process, the words without substantive meaning, such as stop words, modal particles, etc., can be removed first. Then, the hot words and cold words are removed, wherein the hot words and cold words can be judged by the percentile distribution of the number of users using each word. For example, the words that are mentioned by 80% or more of the users are "hot words", and the words with long tail and very few users are "cold words". Finally, the words with smaller weights are removed, and finally the effective keywords are obtained, wherein the words with smaller weights cannot represent the user's behavior, and the weights of the words can be obtained through TFIDF or TEXTRANK.

[0054] In an embodiment of the present application, when clustering the extracted keywords, that is, when classifying the keywords, it can be achieved through the following two steps: 1) inputting the extracted keywords into a pre-trained word2vec model for processing to obtain corresponding word vectors; 2) based on a clustering algorithm, processing the word vectors of the extracted keywords to obtain multiple word classes.

[0055] In the embodiment of the present application, relevant training data can be used to pre-train the word2vec model offline, and if there are new words, they can be directly inferred and applied online.

[0056] In an embodiment of the present application, a k-means clustering algorithm can be used to process the word vectors of the extracted keywords to obtain multiple word classes.

[0057] In an embodiment of the present application, when obtaining the word class features of each word class, it can be achieved through the following three steps: 1) generating comprehensive features of each keyword under each word class based on the user behavior features corresponding to each keyword under each word class in the user text data; 2) generating word weights of each keyword under each word class based on the user behavior corresponding to each keyword under each word class in the user text data; 3) for each word class, performing weighted summation of the comprehensive features of each keyword under the word class and the corresponding word weight to obtain the word class features of the word class.

[0058] In the embodiment of the present application, based on the user behavior features corresponding to the keywords under each word class in the user text data, generating the comprehensive features of the keywords under each word class may specifically include the following steps:

[0059] The user behavior features corresponding to each keyword under each word class in the user text data are processed to obtain the weights of each user behavior feature corresponding to each keyword under each word class, and the user behavior features corresponding to each keyword under each word class are weighted and summed with the corresponding weights to obtain the comprehensive features of each keyword under each word class, wherein the user behavior features include: number of users, number of days, number of times and duration.

[0060] In practical applications, the PCA algorithm can be used to process the user behavior features corresponding to the keywords to obtain the weights of the user behavior features corresponding to the keywords.

[0061] In the embodiment of the present application, according to the user behavior corresponding to each keyword under each word class in the user text data, the word weight of each keyword under each word class is generated, which may specifically include the following steps:

[0062] The user behaviors corresponding to the keywords under each word class in the user text data and the preset multiple portrait tags are processed to obtain the behavior preferences of the portrait tags corresponding to the keywords under each word class, and the behavior preferences corresponding to the keywords are determined as the word weights of the corresponding keywords.

[0063] In practical applications, the attention mechanism can be used to process user behaviors corresponding to keywords and multiple preset portrait tags to obtain the behavioral preferences of the portrait tags corresponding to the keywords.

[0064] In one example, taking word class 1 as an example, if word class 1 includes: keyword 1, keyword 2 and keyword 3, then the word class feature of word class 1 = ∑ (comprehensive feature of keyword i * word weight of keyword i), i = (1, 2, 3); wherein the specific processing process is as follows:

[0065] 1) Generate comprehensive features of keyword 1, comprehensive features of keyword 2, and comprehensive features of keyword 3

[0066] User behavior generates words, and behavior features include the number of users, days, times, duration, etc. Taking keyword 1 as an example, these four types of features of keyword 1 are input into PCA principal component analysis, and the weights of these four types of features of keyword 1 are calculated by using the contribution of the principal component, the linear combination relationship between the principal component factor and various original features, and then fused (i.e., the features and weights are weighted summed) and assigned to keyword 1 to obtain the comprehensive features of keyword 1. Similarly, the comprehensive features of keyword 2 and keyword 3 can be generated.

[0067] 2) Generate the word weight of keyword 1, the word weight of keyword 2, and the word weight of keyword 3

[0068] A word class has multiple words corresponding to multiple user behaviors. Taking keyword 1 as an example, the user behavior corresponding to keyword 1 is combined with the behavior preference of the portrait label and the behavior preference is used as the word weight of keyword 1 according to the attention mechanism. Similarly, the word weight of keyword 2 and keyword 3 can be generated.

[0069] 3) Generate word class features of word class 1

[0070] The word class feature of word class 1 = comprehensive feature of keyword 1 * word weight of keyword 1 + comprehensive feature of keyword 2 * word weight of keyword 2 + comprehensive feature of keyword 3 * word weight of keyword 3.

[0071] It can be seen that in the embodiment of the present application, replacing a single keyword with a word class as an input feature can reduce the head effect of hot scenes or popular words, thereby improving the stability of input and the contribution of the long-tail information dimension. In addition, the weight of the keyword depends not only on the frequency, but also on the duration and active days, and the weight is extracted by PCA, and the final comprehensive weighting is obtained to obtain the basic features, which can obtain higher information content than pure frequency information input.

[0072] In step 102, a word class feature vector is generated for each user based on the generated word class features and the relationship between each user and the word class, wherein the word class feature vector is an N-dimensional vector, where N is the total number of word classes.

[0073] In one example, taking a single user A as an example, multiple keywords are extracted from the text data generated by user A, and the multiple keywords belong to word class 1, word class 2 and word class 3. Assuming that 100 word classes are generated in step 101: word class 1, word class 2, ..., word class 100, the word class feature vector of user A is a 100-dimensional vector, which can be specifically (word class feature of word class 1, word class feature of word class 2, word class feature of word class 3, 0, ..., 0) or (word class feature of word class 1, word class feature of word class 2, word class feature of word class 3, default, ..., default).

[0074] In an example, the user group to be classified includes 200 users, and the word class feature vectors of the 200 users, that is, 200 word class feature vectors, will be generated through step 102 .

[0075] In step 103, the probability distribution of each user with respect to each preset topic is obtained based on the word class feature vector of each user.

[0076] In an embodiment of the present application, the word class feature vector of each user can be input into a preset probabilistic topic model for processing to obtain the probability distribution of each user on each topic.

[0077] In an embodiment of the present application, the probabilistic topic model may be an LDA topic model, the input of which is a word class feature vector of a single user, and the output of which is a probability distribution of each topic for the single user.

[0078] In the embodiment of the present application, relevant training data can be used to pre-train the LDA topic model.

[0079] In one example, the output of the LDA topic model corresponds to six topics: Topic 1, Topic 2, Topic 3, Topic 4, Topic 5, and Topic 6. The user group to be classified includes 200 users, and the probability distribution of these 200 users on these six topics is obtained through step 103. Taking one user as an example, the probability distribution of the user on these six topics is [0.1, 0.2, 0.05, 0.5, 0.1, 0.05].

[0080] In step 104, for each user, the probability distribution of the user regarding the preset business scenarios is generated based on the mapping relationship between the preset word classes and the preset business scenarios, the probability distribution of the word classes under the preset topics, and the probability distribution of the user regarding the preset topics.

[0081] In the embodiment of the present application, the probability of a single user belonging to each business scenario can be calculated and output based on the mapping relationship between topic<-->word class<-->business scenario, thereby obtaining the cognitive output of the user's core business scenario demands.

[0082] In actual applications, business scenarios may include: maternal and child care, early childhood education, shopping, travel, navigation, job hunting, etc.

[0083] In the embodiment of the present application, the probability distribution of the word classes under the preset topics may be: the conjugate matrix of the word classes represented by the probability topic model in step 103 and the preset topics.

[0084] In an implementation manner provided in the present application, the above step 104 may specifically include the following steps (not shown in the figure): step 1041, step 1042 and step 1043, wherein:

[0085] In step 1041, for each user, the probability distribution of the word class under each preset topic is multiplied by the probability distribution of the user with respect to each preset topic to obtain the probability distribution of each word class under each preset topic for the user.

[0086] In step 1042, based on the mapping relationship between the preset word classes and the preset business scenarios, and the probability distribution of the word classes under the preset topics of the user, the probability distribution of the preset topics under the preset business scenarios of the user is generated.

[0087] In step 1043, the probability distribution of the user regarding each preset topic under each preset business scenario is multiplied by the probability distribution of the user regarding each preset topic to generate the probability distribution of the user regarding each preset business scenario.

[0088] In an example, the detailed process can be as follows:

[0089] 1) Pre-establish the mapping relationship between word classes and business scenarios

[0090] This can be determined by pre-training and preliminary configuration. The business scenario definition and description are specified by business personnel. Then, based on the scenario description defined by the business, according to the graph relationship embedding and the average similarity of the word set in the descriptive words and word classes, the word class most similar to the business scenario is automatically given, forming a relationship suggestion of business scenario <--> word class. Then, business personnel can perform manual inspection and correction if necessary to improve the efficiency of establishing the mapping relationship; for example: word class 1 (lottery; financial management test; bank; loan) <--> financial management scenario.

[0091] 2) Calculate the probability of business scenarios

[0092] like Figure 2 The overall computing link is mainly composed of three computing components: Figure 2 The left one in the figure is the pre-built mapping relationship between word classes and business scenarios, which serves as a dimension table; Figure 2The middle of is the conjugate matrix result of word classes and topics output by LDA in step 103. The vertical direction represents word classes, the row number represents the word class number, and the horizontal direction represents the result topic, the column number represents the topic number, and the value represents the multinomial distribution of the topic and the generated word class, as the result dimension table; Figure 2 The right one in the figure is the probability distribution of a single user on each topic output by LDA.

[0093] First, the probability distribution of each word class under each topic for a single user is calculated by dividing the probability distribution of each user on each topic by the conjugate matrix of each word class and each topic represented by the probabilistic topic model.

[0094] For example, the probability of a single user A on topic 1 is 0.1, and the multinomial distribution of topic 1 on each word class is [0.3, 0.1, 0.25, 0.15, 0.2]. Then the probability distribution of user A on each word class under topic 1 is 0.1*[0.3, 0.1, 0.25, 0.15, 0.2] = [0.03, 0.01, 0.025, 0.015, 0.02]. The same can be said for other topics. Finally, the distribution matrix of user A's topic and generated word classes is obtained, as shown in Figure 3 shown.

[0095] Afterwards, according to the word class dimension, the mapping relationship between word classes and business scenarios is combined with Figure 3 The results shown in the figure are integrated and summarized into the probability distribution of each business scenario for a single user. Figure 4 As shown, first map the word classes to the corresponding business scenario numbers based on the mapping relationship between word classes and business scenarios. Figure 4 As shown in the left one in , the columns are then grouped by business scenario and the probability distribution of each topic under each business scenario is obtained as follows Figure 4 As shown in the middle of , finally, according to the user's topic distribution probability vector (transposed into a column vector), the value of each row is weighted and summed to obtain the user's probability distribution for each business scenario as shown below Figure 4 As shown in the right figure.

[0096] As can be seen from the above embodiments, in this embodiment, for the user group to be classified, the keywords in the user text data generated by the user group are extracted, the keywords are clustered to obtain multiple word classes, and the word class features of each word class are generated. After that, the word class feature vector of each user is generated, and the probability distribution of each user on each topic is obtained according to the word class feature vector of each user; finally, according to the mapping relationship between the topic, word class and business scenario, and the probability distribution of each user on each topic, the probability distribution of each user in the user group on each business scenario is generated. Compared with the prior art, in the embodiment of the present application, the keywords are classified, and the word class is used to replace a single word as an input feature to classify the user group, which can reduce the head effect of the hot scene or popular words, improve the stability of the input and the contribution of the long-tail information dimension, thereby improving the accuracy of user classification. In addition, taking the word class as the feature input unit, establishing a low-dimensional mapping of the topic, word class and business scenario, and classifying the user group, it can be not only from the user perspective, but also from the business scenario perspective when finally recognizing the user, and it can be switched from the business scenario perspective, which is efficient, stable and converges to the business direction to automatically describe the user, and quickly help product operations recognize users.

[0097] In another embodiment provided by the present application, in the embodiment of the present application, the top concerned keywords and business scenario needs of the crowd can be output, and copywriting suggestions can be output, so as to assist the user dimension to the business dimension to enter the cognition and implementation, such as Figure 5 As shown, the user classification method provided in the embodiment of the present application can also be used in Figure 1 After step 104 of the illustrated embodiment, the following steps are added: step 501, step 502, step 503 and step 504, wherein:

[0098] In step 501, for each word class under each topic, the word weight of each keyword under the word class is multiplied by the probability distribution of the corresponding word class under the topic to obtain the probability distribution of each keyword under the word class under the topic.

[0099] In step 502, the user coverage and click-through rate of each keyword under the topic are obtained.

[0100] In step 503, for each keyword under the topic, the probability, user coverage and click rate of the keyword under the topic are multiplied to obtain the recommendation coefficient of the keyword under the topic.

[0101] In step 504, business copy suggestions containing corresponding keywords are output according to the descending order of the recommendation coefficients of the keywords.

[0102] In an example, the user classification results are shown as follows:

[0103] 1) Core scenarios: childcare, maternal and infant care, and early childhood education;

[0104] 2) Automatic description of the crowd (high TGI high proportion):

[0105] Female (66.14%, 132)

[0106] 24-27 years old (67.52%, 1157)

[0107] Married (79.81%, 249), with children (64.03%, 127),

[0108] Prefer mobile phones priced between 700-999 yuan (33.74%, 247) and 1000-1499 yuan (27.60%, 112);

[0109] 3) Top keywords in top scenarios are displayed through word cloud diagrams, such as Figure 6 shown.

[0110] It can be seen that in the embodiments of the present application, the top keywords + business scenario needs can be output based on the crowd, and copywriting suggestions can be output to assist in the entry and cognition from the user dimension to the business dimension from implementation.

[0111] In another embodiment provided by the present application, in the embodiment of the present application, after the crowd is clustered, the key difference description of the crowd can be automatically output according to the importance (information contribution and TGI) of the keywords and portrait tags, thereby improving the efficiency of user cognition of product operation, such as Figure 7 As shown, the user classification method provided in the embodiment of the present application can also be used in Figure 1 After step 104 of the illustrated embodiment, the following steps are added: step 701, step 702 and step 703, wherein:

[0112] In step 701, a preset portrait tag set is obtained, wherein the portrait tag set includes a plurality of portrait tags.

[0113] In step 702, based on the probability distribution of each business scenario preset by the user, the correlation between each business scenario and each portrait tag in the portrait tag set is calculated.

[0114] In the embodiment of the present application, when calculating the correlation, if the label is numerical (such as age), the Pearson correlation coefficient is used, and if the label is discrete (such as gender), the rank correlation coefficient is used.

[0115] In step 703, the portrait tag with a higher correlation is used as the crowd description information of the user.

[0116] In the embodiment of the present application, a higher correlation means: a correlation greater than a preset threshold; a portrait tag with a higher correlation means: a portrait tag whose correlation is greater than a preset threshold.

[0117] It can be seen that in the embodiment of the present application, after the crowd is clustered, the key difference description of the crowd can be automatically output according to the importance of keywords and portrait tags (information contribution and TGI), thereby improving the efficiency of user cognition of product operation.

[0118] In another embodiment provided by the present application, in the embodiment of the present application, when it is necessary to find users who follow the rights and interests of a new product, a scene association search can be performed based on the graph entity, and the expanded words are further located in the corresponding word class to find new rights and interests seed users, such as Figure 8 As shown, the user classification method provided in the embodiment of the present application can also be used in Figure 1 After step 104 of the illustrated embodiment, the following steps are added: step 801, step 802, step 803 and step 804, wherein:

[0119] In step 801, for the new rights and interests that are launched, relevant keywords of the new rights and interests are obtained, and the relevant keywords are classified to obtain corresponding relevant word classes.

[0120] In the embodiment of the present application, the new benefit may be an additional service given to the user, such as a mobile phone case, a film, or other value-added products.

[0121] In an embodiment of the present application, when obtaining relevant keywords for new rights and interests, the new rights and interests keywords for business products entered by business personnel can be used as relevant keywords, and relevant keywords can also be obtained through word expansion, for example, by obtaining relevant associated words based on scene associations and entity relationships through a knowledge graph.

[0122] In step 802, a word class feature vector of the new equity is generated based on the generated word class features and related word classes, and a probability distribution of the new equity with respect to each topic is obtained based on the word class feature vector of the new equity.

[0123] In the embodiment of the present application, the word class feature vector of the new rights and interests can be input into the probabilistic topic model for processing to obtain the probability distribution of the new rights and interests with respect to each topic.

[0124] In step 803, based on the probability distribution of the new rights and interests with respect to each topic, the head topic to which the new rights and interests belong is determined, wherein the probability value of the head topic is the largest.

[0125] In step 804, based on the probability distribution of users on each topic, a user group with the same head topic as the new benefit is determined among existing users, and the user group is determined as the seed user of the new benefit.

[0126] It can be seen that in the embodiment of the present application, a function of selecting target groups for new rights and interests is added. When a business has new rights and interests but no corresponding portrait tags, this function can be selected. On the premise of recognizing new users, it can also assist new businesses in finding initial more matching seed users and assist in initializing push and reach services for new business rights and interests.

[0127] It should be noted that the user classification method provided in the embodiment of the present application can be executed by a user classification device, or a control module in the user classification device for executing the loaded user classification method. In the embodiment of the present application, the user classification device provided in the embodiment of the present application is described by taking the user classification device executing the loaded user classification method as an example.

[0128] Fig. 9 is a structural block diagram of a user classification device provided in an embodiment of the present application, such as Fig. 9 As shown, the user classification device 900 may include: a first acquisition module 901, a first generation module 902, a second acquisition module 903 and a second generation module 904, wherein:

[0129] A first acquisition module 901 is used to acquire a word class feature of each word class based on user text data, wherein the word class is obtained based on keyword clustering in the user text data;

[0130] A first generating module 902 is used to generate a word class feature vector for each user according to the generated word class features and the relationship between each user and the word class, wherein the word class feature vector is an N-dimensional vector, where N is the total number of word classes;

[0131] The second acquisition module 903 is used to obtain the probability distribution of each user with respect to each preset topic according to the word class feature vector of each user;

[0132] The second generation module 904 is used to generate, for each user, the probability distribution of the user regarding the preset business scenarios based on the mapping relationship between the preset word classes and the preset business scenarios, the probability distribution of the word classes under the preset topics, and the probability distribution of the user regarding the preset topics.

[0133] As can be seen from the above embodiments, in this embodiment, for the user group to be classified, the keywords in the user text data generated by the user group are extracted, the keywords are clustered to obtain multiple word classes, and the word class features of each word class are generated. After that, the word class feature vector of each user is generated, and the probability distribution of each user on each topic is obtained according to the word class feature vector of each user; finally, according to the mapping relationship between the topic, word class and business scenario, and the probability distribution of each user on each topic, the probability distribution of each user in the user group on each business scenario is generated. Compared with the prior art, in the embodiment of the present application, the keywords are classified, and the word class is used to replace a single word as an input feature to classify the user group, which can reduce the head effect of the hot scene or popular words, improve the stability of the input and the contribution of the long-tail information dimension, thereby improving the accuracy of user classification. In addition, taking the word class as the feature input unit, establishing a low-dimensional mapping of the topic, word class and business scenario, and classifying the user group, it can be not only from the user perspective, but also from the business scenario perspective when finally recognizing the user, and it can be switched from the business scenario perspective, which is efficient, stable and converges to the business direction to automatically describe the user, and quickly help product operations recognize users.

[0134] Optionally, as an embodiment, the second generating module 904 may include:

[0135] The first generating submodule is used to multiply the probability distribution of the word class under each preset topic by the probability distribution of the user on each preset topic for each user, so as to obtain the probability distribution of each word class under each preset topic for the user;

[0136] The second generating submodule is used to generate the probability distribution of the user's preset topics under the preset business scenarios according to the mapping relationship between the preset word classes and the preset business scenarios and the probability distribution of the user's word classes under the preset topics;

[0137] The third generating submodule is used to multiply the probability distribution of the user regarding each preset topic under each preset business scenario by the probability distribution of the user regarding each preset topic, so as to generate the probability distribution of the user regarding each preset business scenario.

[0138] Optionally, as an embodiment, the first obtaining module 901 may include:

[0139] The fourth generation submodule is used to process the user behavior features corresponding to each keyword under each word class in the user text data, obtain the weight of each user behavior feature corresponding to each keyword under each word class, and perform weighted summation on each user behavior feature corresponding to each keyword under each word class and the corresponding weight to obtain the comprehensive features of each keyword under each word class, wherein the user behavior features include: number of users, number of days, number of times and duration;

[0140] The fifth generation submodule is used to process the user behavior corresponding to each keyword under each word class in the user text data and the preset multiple portrait tags to obtain the behavior preference of the portrait tag corresponding to each keyword under each word class, and determine the behavior preference corresponding to each keyword as the word weight of the corresponding keyword;

[0141] The sixth generation submodule is used to perform weighted summation of the comprehensive features of each keyword under each word class and the corresponding word weight for each word class to obtain the word class feature of the word class.

[0142] Optionally, as an embodiment, the user classification device 900 may further include:

[0143] A third generation module is used for, for each word class under each topic, multiplying the word weight of each keyword under the word class by the probability distribution of the corresponding word class under the topic to obtain the probability distribution of each keyword under the word class under the topic;

[0144] The third acquisition module is used to obtain the user coverage and click rate of each keyword under the topic;

[0145] A fourth generating module is used for multiplying the probability, user coverage and click rate of each keyword under the theme to obtain a recommendation coefficient of the keyword under the theme;

[0146] The fifth generation module is used to output business copy suggestions containing corresponding keywords in descending order of recommendation coefficients of the keywords.

[0147] Optionally, as an embodiment, the user classification device 900 may further include:

[0148] A fourth acquisition module is used to acquire a preset portrait tag set, wherein the portrait tag set includes a plurality of portrait tags;

[0149] A calculation module, used to calculate the correlation between each business scenario and each portrait tag in the portrait tag set according to the probability distribution of each business scenario preset by the user;

[0150] The determination module is used to use the portrait tags with higher relevance as the population description information of the user.

[0151] The user classification device in the embodiment of the present application can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, a wearable device, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA), etc. The non-mobile electronic device can be a server, a network attached storage (NAS), a personal computer (PC), a television (TV), a teller machine or a self-service machine, etc., which is not specifically limited in the embodiment of the present application.

[0152] The spatial position search device in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0153] The user classification device provided in the embodiment of the present application can achieve Figure 1 To avoid repetition, the various processes implemented in the method embodiment will not be described again here.

[0154] Alternatively, if Fig.10 As shown, an embodiment of the present application also provides an electronic device 1000, including a processor 1001, a memory 1002, and a program or instruction stored in the memory 1002 and executable on the processor 1001. When the program or instruction is executed by the processor 1001, each process of the above-mentioned user classification method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0155] It should be noted that the electronic devices in the embodiments of the present application include the mobile electronic devices and non-mobile electronic devices mentioned above.

[0156] Fig.11 The hardware structure diagram of an electronic device for implementing an embodiment of the present application is shown in FIG. The electronic device 1100 includes, but is not limited to, a radio frequency unit 1101, a network module 1102, an audio output unit 1103, an input unit 1104, a sensor 1105, a display unit 1106, a user input unit 1107, an interface unit 1108, a memory 1109, and a processor 1110.

[0157] Those skilled in the art will appreciate that the electronic device 1100 may also include a power source (such as a battery) for supplying power to each component, and the power source may be logically connected to the processor 1110 through a power management system, thereby implementing functions such as managing charging, discharging, and power consumption management through the power management system. Fig.11 The electronic device structure shown in the figure does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently, which will not be described in detail here.

[0158] Among them, processor 1110 is used to obtain the word class features of each word class based on user text data, wherein the word class is obtained based on keyword clustering in the user text data; generate a word class feature vector for each user based on the generated word class features and the relationship between each user and the word class, wherein the word class feature vector is an N-dimensional vector, and N is the total number of word classes; obtain the probability distribution of each user regarding preset topics based on the word class feature vector of each user; for each user, generate the probability distribution of the user regarding preset business scenarios based on the mapping relationship between the preset word class and the preset business scenarios, the probability distribution of the word class under the preset topics, and the probability distribution of the user regarding the preset topics.

[0159] It can be seen that in the embodiment of the present application, by classifying keywords and replacing individual words with word classes as input features to classify user groups, the head effect of hot scenes or popular words can be reduced, the stability of input and the contribution of long-tail information dimensions can be improved, thereby improving the accuracy of user classification. In addition, by using word classes as feature input units, establishing a low-dimensional mapping of topics, word classes and business scenarios, and classifying user groups, it is possible to switch from not only the user perspective but also the business scenario perspective when finally recognizing users, efficiently and stably and converging to the business direction to automatically describe users, and quickly help product operations recognize users.

[0160] Optionally, as an embodiment, the processor 1110 is further configured to, for each user, multiply the probability distribution of the word class under each preset topic by the probability distribution of the user with respect to each preset topic, to obtain the probability distribution of each word class under each preset topic of the user;

[0161] Generate the probability distribution of the user's preset topics under the preset business scenarios according to the mapping relationship between the preset word classes and the preset business scenarios, and the probability distribution of the user's preset word classes under the preset topics;

[0162] The probability distribution of the user regarding each preset topic under each preset business scenario is multiplied by the probability distribution of the user regarding each preset topic to generate the probability distribution of the user regarding each preset business scenario.

[0163] Optionally, as an embodiment, the processor 1110 is further configured to process the user behavior features corresponding to each keyword under each word class in the user text data, obtain the weights of each user behavior feature corresponding to each keyword under each word class, perform weighted summation on each user behavior feature corresponding to each keyword under each word class and the corresponding weight, and obtain the comprehensive features of each keyword under each word class, wherein the user behavior features include: number of users, number of days, number of times and duration;

[0164] Processing the user behaviors corresponding to each keyword under each word class in the user text data and a plurality of preset portrait tags to obtain the behavior preferences of the portrait tags corresponding to each keyword under each word class, and determining the behavior preferences corresponding to each keyword as the word weight of the corresponding keyword;

[0165] For each word class, the comprehensive features of each keyword under the word class and the corresponding word weight are weighted and summed to obtain the word class feature of the word class.

[0166] Optionally, as an embodiment, the processor 1110 is further configured to, for each word class under each topic, multiply the word weight of each keyword under the word class by the probability distribution of the corresponding word class under the topic to obtain the probability distribution of each keyword under the word class under the topic;

[0167] Obtain user coverage and click-through rate of each keyword under the topic;

[0168] For each keyword under the topic, multiply the probability, user coverage and click-through rate of the keyword under the topic to obtain a recommendation coefficient of the keyword under the topic;

[0169] Output business copy suggestions containing corresponding keywords in descending order of the recommendation coefficients of the keywords.

[0170] Optionally, as an embodiment, the processor 1110 is further configured to obtain a preset portrait tag set, where the portrait tag set includes a plurality of portrait tags;

[0171] Calculate the correlation between each business scenario and each portrait tag in the portrait tag set according to the probability distribution of each business scenario preset by the user;

[0172] The portrait tags with higher relevance are used as the population description information of the user.

[0173] It should be understood that in the embodiment of the present application, the input unit 1104 may include a graphics processor (Graphics Processing Unit, GPU) 11041 and a microphone 11042, and the graphics processor 11041 processes the image data of the static picture or video obtained by the image capture device (such as a camera) in the video capture mode or the image capture mode. The display unit 1106 may include a display panel 11061, and the display panel 11061 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc. The user input unit 1107 includes a touch panel 11071 and other input devices 11072. The touch panel 11071 is also called a touch screen. The touch panel 11071 may include two parts: a touch detection device and a touch controller. Other input devices 11072 may include, but are not limited to, a physical keyboard, a function key (such as a volume control button, a switch button, etc.), a trackball, a mouse, and a joystick, which will not be repeated here. The memory 1109 can be used to store software programs and various data, including but not limited to applications and operating systems. The processor 1110 may integrate an application processor and a modem processor, wherein the application processor mainly processes an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the modem processor may not be integrated into the processor 1110.

[0174] The electronic device 1100 can implement various processes implemented by the electronic devices in the aforementioned embodiments, which will not be described again here to avoid repetition.

[0175] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of the above-mentioned user classification method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0176] The processor is a processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0177] The embodiment of the present application further provides a chip, the chip includes a processor and a communication interface, the communication interface is coupled to the processor, the processor is used to run a program or instruction, implement the various processes of the above-mentioned user classification method embodiment, and can achieve the same technical effect, to avoid repetition, it is not repeated here. It should be understood that the chip mentioned in the embodiment of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0178] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0179] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0180] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A user classification method, characterized in that: The method comprises: According to the user text data, obtaining a word class feature of each word class, wherein the word class is obtained based on keyword clustering in the user text data; Generate a word class feature vector for each user based on the generated word class features and the relationship between each user and the word class, wherein the word class feature vector is an N-dimensional vector, where N is the total number of word classes; According to each user's word class feature vector, obtain each user's probability distribution on each preset topic; For each user, based on the mapping relationship between preset word classes and preset business scenarios, the probability distribution of word classes under preset topics, and the probability distribution of the user with respect to preset topics, generate the probability distribution of the user with respect to preset business scenarios; For each user, according to the mapping relationship between preset word classes and preset business scenarios, the probability distribution of word classes under preset topics, and the probability distribution of the user with respect to preset topics, the probability distribution of the user with respect to preset business scenarios is generated, including: For each user, multiply the probability distribution of the word class under each preset topic by the probability distribution of the user on each preset topic to obtain the probability distribution of each word class under each preset topic for the user; Generate the probability distribution of the user's preset topics under the preset business scenarios according to the mapping relationship between the preset word classes and the preset business scenarios, and the probability distribution of the user's preset word classes under the preset topics; The probability distribution of the user regarding each preset topic under each preset business scenario is multiplied by the probability distribution of the user regarding each preset topic to generate the probability distribution of the user regarding each preset business scenario.

2. The method according to claim 1, characterized in that The step of obtaining the word class feature of each word class according to the user text data includes: Processing the user behavior features corresponding to each keyword under each word class in the user text data to obtain the weights of each user behavior feature corresponding to each keyword under each word class, performing weighted summation on each user behavior feature corresponding to each keyword under each word class and the corresponding weight to obtain the comprehensive features of each keyword under each word class, wherein the user behavior features include: number of users, number of days, number of times and duration; Processing the user behaviors corresponding to each keyword under each word class in the user text data and the preset multiple portrait tags to obtain the behavior preferences of the portrait tags corresponding to each keyword under each word class, and determining the behavior preferences corresponding to each keyword as the word weight of the corresponding keyword; For each word class, the comprehensive features of each keyword under the word class and the corresponding word weight are weighted and summed to obtain the word class feature of the word class.

3. The method according to claim 2, characterized in that After generating the probability distribution of the user for each preset business scenario, the method further includes: For each word class under each topic, multiply the word weight of each keyword under the word class by the probability distribution of the corresponding word class under the topic to obtain the probability distribution of each keyword under the word class under the topic; Obtain user coverage and click-through rate of each keyword under the topic; For each keyword under the topic, multiply the probability, user coverage and click-through rate of the keyword under the topic to obtain a recommendation coefficient of the keyword under the topic; Output business copy suggestions containing corresponding keywords in descending order of recommendation coefficients of keywords.

4. The method according to claim 1, characterized in that: After generating the probability distribution of the user for each preset business scenario, the method further includes: Obtain a preset portrait tag set, wherein the portrait tag set includes multiple portrait tags; Calculate the correlation between each business scenario and each portrait tag in the portrait tag set according to the probability distribution of each business scenario preset by the user; The portrait tags with higher relevance are used as the population description information of the user.

5. A user classification device, characterized in that: The device comprises: A first acquisition module, configured to acquire a word class feature of each word class according to user text data, wherein the word class is obtained by clustering keywords in the user text data; A first generating module, configured to generate a word class feature vector for each user based on the generated word class features and the relationship between each user and the word class, wherein the word class feature vector is an N-dimensional vector, where N is the total number of word classes; The second acquisition module is used to obtain the probability distribution of each user on each preset topic according to the word class feature vector of each user; A second generating module is used to generate, for each user, a probability distribution of the user with respect to each preset business scenario according to a mapping relationship between preset word classes and preset business scenarios, a probability distribution of word classes under each preset topic, and a probability distribution of the user with respect to each preset topic; The second generation module comprises: The first generating submodule is used to multiply the probability distribution of the word class under each preset topic by the probability distribution of the user on each preset topic for each user, so as to obtain the probability distribution of each word class under each preset topic for the user; The second generating submodule is used to generate the probability distribution of the user's preset topics under the preset business scenarios according to the mapping relationship between the preset word classes and the preset business scenarios and the probability distribution of the user's word classes under the preset topics; The third generating submodule is used to multiply the probability distribution of the user regarding each preset topic under each preset business scenario by the probability distribution of the user regarding each preset topic, so as to generate the probability distribution of the user regarding each preset business scenario.

6. The device according to claim 5, characterized in that The first acquisition module includes: The fourth generation submodule is used to process the user behavior features corresponding to each keyword under each word class in the user text data, obtain the weight of each user behavior feature corresponding to each keyword under each word class, and perform weighted summation on each user behavior feature corresponding to each keyword under each word class and the corresponding weight to obtain the comprehensive features of each keyword under each word class, wherein the user behavior features include: number of users, number of days, number of times and duration; The fifth generation submodule is used to process the user behavior corresponding to each keyword under each word class in the user text data and the preset multiple portrait tags to obtain the behavior preference of the portrait tag corresponding to each keyword under each word class, and determine the behavior preference corresponding to each keyword as the word weight of the corresponding keyword; The sixth generation submodule is used to perform weighted summation of the comprehensive features of each keyword under each word class and the corresponding word weight for each word class to obtain the word class feature of the word class.

7. The device according to claim 6, characterized in that The device also includes: A third generation module is used for multiplying the word weight of each keyword under each word class by the probability distribution of the corresponding word class under the topic for each word class under each topic, so as to obtain the probability distribution of each keyword under the word class under the topic; The third acquisition module is used to obtain the user coverage and click rate of each keyword under the topic; A fourth generating module, for each keyword under the theme, multiplying the probability, user coverage and click rate of the keyword under the theme to obtain a recommendation coefficient of the keyword under the theme; The fifth generation module is used to output business copy suggestions containing corresponding keywords in descending order of recommendation coefficients of the keywords.

8. The device according to claim 5, characterized in that The device also includes: A fourth acquisition module is used to acquire a preset portrait tag set, wherein the portrait tag set includes a plurality of portrait tags; A calculation module, used to calculate the correlation between each business scenario and each portrait tag in the portrait tag set according to the probability distribution of each business scenario preset by the user; The determination module is used to use the portrait tags with higher relevance as the population description information of the user.

9. An electronic device, characterized in that: The electronic device comprises a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of the user classification method according to any one of claims 1 to 4 are implemented.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the steps of the user classification method according to any one of claims 1 to 4 are implemented.

Images