Hierarchical incremental label propagation method and device based on probability distribution

Abstract

The application relates to a hierarchical incremental label propagation method and device based on a probability distribution. The method comprises the following steps: constructing a relationship network through relationships among a plurality of users; assigning user labels to a first part of users in the relationship network according to a label strategy; generating a label dimension set of a second part of users in the relationship network through a probability distribution of the user labels of the first part of users according to a hierarchical label propagation algorithm; and determining user labels of the second part of users according to the label dimension set. The hierarchical incremental label propagation method and device based on the probability distribution, the electronic equipment and the computer readable medium can quickly and accurately determine user labels of users without labels in actual application, meet the demand of calculation and analysis, and compared with a label assignment mode in the prior art, the mode in the application improves the calculation speed and reduces the occupation of content and calculation resources.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of computer information processing, in particular, to a hierarchical incremental label propagation method based on probability distribution, device, electronic equipment and computer readable medium. BACKGROUND

[0002] Machine learning has now been greatly developed in various artificial intelligence research fields. For general application of machine learning models, a user first determines a machine learning model of a certain category or algorithm, then inputs specific data according to a specific problem the user wants to solve, the machine learning model establishes a specific task, and then trains the machine learning through specific data. After the training is completed, a machine learning model suitable for a specific task is obtained. In general, the same machine learning model algorithm is used, but the machine learning models trained by different data are completely different.

[0003] In general, machine learning models need to learn positive samples and negative samples. However, for many fields, most users belong to unlabeled samples. For example, in the shopping field, whether a user likes a commodity (positive sample) can be determined by the user's purchase habits; whether the user dislikes a commodity (negative sample) can also be determined by the user's evaluation content, but whether most of the users like the commodity is unknown (unlabeled sample). A large number of unlabeled samples will affect the precision and accuracy of the machine learning model during training. How to accurately and quickly assign sample labels to unlabeled users and then train the machine learning model is a key problem in the machine learning model training process.

[0004] Therefore, a new hierarchical incremental label propagation method based on probability distribution, device, electronic equipment and computer readable medium are needed.

[0005] The above information disclosed in the background section is only intended to strengthen the understanding of the background of the present application, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] Therefore, the present application provides a hierarchical incremental label propagation method based on probability distribution, device, electronic equipment and computer readable medium, which can quickly and accurately determine the user label of the user without label in actual application, meet the demand of calculation and analysis, compared with the label assignment method in the prior art, the method in the present application improves the calculation speed and reduces the occupation of content and computing resources.

[0007] Other characteristics and advantages of the present application will become apparent from the following detailed description, or will be learned by practice of the present application.

[0008] According to an aspect of the present application, a hierarchical incremental label propagation method based on probability distribution is provided, which comprises: constructing a relationship network through relationships among a plurality of users; assigning user labels to a first part of users in the relationship network according to a label strategy; generating a label dimension set of a second part of users in the relationship network through the user labels of the first part of users according to a hierarchical label propagation algorithm; determining user labels for the second part of users according to the label dimension set.

[0009] Optionally, assigning user labels to the first part of users in the relationship network according to the label strategy comprises: determining a target and a dimension of the user labels to be assigned; generating the label strategy according to the target and the dimension; and assigning user labels to the first part of users according to the label strategy and historical data of the plurality of users in the relationship network.

[0010] Optionally, assigning user labels to the first part of users according to the label strategy and the historical data of the plurality of users in the relationship network comprises: obtaining the historical data of the plurality of users; extracting target data from the historical data according to the target in the label strategy; and comparing the target data with the label strategy to assign user labels to the first part of users.

[0011] Optionally, comparing the target data with the label strategy to assign user labels to the first part of users comprises: determining a label interval according to the target and the dimension in the label strategy; comparing the target data with the label interval; and assigning labels to the first part of users that meet the label interval.

[0012] Optionally, generating the label dimension set of the second part of users in the relationship network through the probability distribution of the user labels of the first part of users according to the hierarchical label propagation algorithm comprises: setting the first part of users as source nodes in the relationship network; propagating labels corresponding to the source nodes to nodes corresponding to the second part of users in the relationship network in a hierarchical relationship according to a label propagation algorithm and the probability distribution; stopping the propagation when a change parameter of the relationship network meets a threshold value; and generating the label dimension set through the dimension and quantity of the labels of the nodes corresponding to the second part of users.

[0013] Optionally, the label propagation algorithm and the probability distribution are used to propagate the label corresponding to the source node to the nodes corresponding to the second part of users in the relationship network in a hierarchical manner, including: setting the source node as an active node and setting the nodes corresponding to the second part of users as non-active nodes; extracting the active nodes one by one, and passing the label array of the source node to other nodes; other nodes generate their own temporary label of this time according to the probability distribution of the label array of the source node; when the temporary label of this time is different from the last temporary label, the other nodes are set as active nodes.

[0014] Optionally, other nodes generate their own temporary label of this time according to the probability distribution of the label array of the source node, including: other nodes update their own label array according to the label array of the source node; calculate the probability corresponding to each temporary label according to the plurality of temporary labels and the corresponding label quantity in the label data; determine the temporary label of this time for the second part of users according to the probability corresponding to each temporary label.

[0015] Optionally, when the change quantity parameter of the relationship network meets the threshold value, the propagation is stopped, including: calculating the change quantity of active nodes in the relationship network after each propagation; when the change quantity of active nodes is less than the threshold value, the propagation is stopped.

[0016] Optionally, the user label for the second part of users is determined according to the label dimension set, including: obtaining a plurality of labels and the corresponding label quantity in the label set; calculating the probability corresponding to each label according to the plurality of labels and the corresponding label quantity; determining the user label for the second part of users according to the probability corresponding to each label.

[0017] Optionally, the user label for the second part of users is determined according to the probability corresponding to each label, including: extracting the label level corresponding to each label according to the label interval; comparing the sum of the label probability of the current level and the label probability of the previous level according to the label level; when the label probability of the current level is greater than the sum of the label probability of the previous level, the current label is taken as the user label of the second part of users.

[0018] According to an aspect of the present application, a hierarchical incremental label propagation device based on probability distribution is provided, which comprises: a network module for constructing a relationship network through the relationship between a plurality of users; an allocation module for allocating user labels to a first part of users in the relationship network according to a label strategy; a propagation module for generating a label dimension set of a second part of users in the relationship network through the probability distribution of the user labels of the first part of users according to a hierarchical label propagation algorithm; and a label module for determining user labels for the second part of users according to the label dimension set.

[0019] According to an aspect of the present application, an electronic device is provided, the electronic device comprising: one or more processors; a storage storing one or more programs; and the one or more programs, when executed by the one or more processors, causing the one or more processors to implement the method as above.

[0020] According to an aspect of the present application, a computer readable medium having stored thereon a computer program which, when executed by a processor, implements the method as above.

[0021] According to the hierarchical incremental label propagation method based on probability distribution, the device, the electronic device and the computer readable medium of the present application, the relationship network is constructed according to the relationship between the multiple users; the user label is assigned to the first part of users in the relationship network according to the label strategy; the label dimension set of the second part of users in the relationship network is generated according to the probability distribution of the user label of the first part of users according to the hierarchical label propagation algorithm; the way of determining the user label of the second part of users according to the label dimension set can quickly and accurately determine the user label of the user without label in actual application, meet the demand of calculation and analysis, compared with the label assignment mode in the prior art, the way in the present application improves the calculation speed and reduces the occupation of content and computing resources.

[0022] It should be understood that the above general description and the following detailed description are only exemplary and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0023] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which:

[0024] Figure 1 is a flow chart of a hierarchical incremental label propagation method based on probability distribution according to an example embodiment.

[0025] Figure 2 is a schematic diagram of a hierarchical incremental label propagation method based on probability distribution according to another example embodiment.

[0026] Figure 3 is a flow chart of a hierarchical incremental label propagation method based on probability distribution according to another example embodiment.

[0027] Figure 4is a flowchart of a probability distribution based hierarchical incremental label propagation method according to another exemplary embodiment.

[0028] Figure 5 is a block diagram of a probability distribution based hierarchical incremental label propagation apparatus according to an exemplary embodiment.

[0029] Figure 6 is a block diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0030] Example embodiments now will be described more fully hereinafter with reference to the accompanying drawings; however, the example embodiments can be implemented in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the example embodiments to those skilled in the art. Like reference numerals refer to like elements throughout the several views.

[0031] Moreover, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the application. One skilled in the relevant art will recognize, however, that the application can be practiced without one or more of the specific details, or with other methods, components, materials, and so forth. In other instances, well-known structures, devices, implementations, and operations have not been shown or described in detail to avoid obscuring aspects of the application.

[0032] The block diagrams in the drawings show functions and functionality as they can be implemented in computing platforms that perform particular tasks. Not shown are the various elements that can be required to interconnect the blocks such that the functionalities are carried out. For example, one can appreciate that a data store can be employed in some embodiments and other storage means can be used instead of, or in addition to, the data store. Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the application. One skilled in the relevant art will recognize, however, that the application can be practiced without one or more of the specific details, or with other methods, components, materials, and so forth. In other instances, well-known structures, devices, implementations, and operations have not been shown or described in detail to avoid obscuring aspects of the application.

[0033] The flow diagrams in the drawings show example sequences of operations that can be implemented in software, hardware, or a combination thereof. In the context of software, the operations can represent computer-executable instructions stored on one or more computer-readable media that, when executed by one or more processors, perform the recited operations. Generally, computer-executable instructions include routines, subroutines, programs, functions, methods, procedures, modules, software, and / or code segments that perform particular tasks or implement particular abstract data types. The form of code can further be differentiated into source code or object code, where source code is associated with a programming language that can be used to create an executable program, and object code is produced by a compiler from source code. In some embodiments, the order of the operations can be rearranged.

[0034] It should be understood that, although terms such as first, second, third, etc. can be used herein to describe various elements, these elements should not be limited by these terms. These terms are used to distinguish one element from another. Thus, a first element discussed below could be termed a second element without departing from the teachings of the present disclosure. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0035] Those skilled in the art can understand that the drawings are only schematic views of the example embodiments, and the modules or flows in the drawings are not necessarily essential for implementing the present application, and therefore cannot be used to limit the protection scope of the present application.

[0036] The inventors of the present application have found through research and analysis that a label propagation algorithm (LabelPropagation) can be used to assign user labels to unknown samples. The label propagation algorithm is a graph-based semi-supervised learning method, and the basic idea is to use the label information of the labeled nodes to predict the label information of the unlabeled nodes. A graph is built using the relationship between samples, the nodes include labeled and unlabeled data, the edges represent the similarity between two nodes, and the label of a node is transferred to other nodes according to the similarity. The label data is like a source, which can label the unlabeled data. The greater the similarity between nodes, the easier the label is to propagate.

[0037] There are two common label propagation algorithms, LPA and SLPA. LPA is a common label propagation algorithm, in which each node receives the labels of neighboring nodes and takes the most frequently occurring label as its new label. LPA algorithm is more commonly used for community discovery rather than label prediction. Moreover, LPA algorithm is a static propagation, i.e., each node propagates in each round, which consumes a lot of memory. Each node outputs only one label, and the output information is less, which is difficult to support subsequent tasks and can only handle independent labels. SLPA is an improved algorithm of LPA, in which each node has multiple labels, and speaker and listener nodes are randomly selected for alternating propagation each time. SLPA algorithm can handle overlapping multi-label propagation, but cannot handle progressive hierarchical labels. Moreover, in the calculation process, the label is statically propagated, which consumes memory. The node output is multiple labels, but there is no primary and secondary relationship between the multiple labels.

[0038] In view of the technical defects in the prior art, the present application proposes a hierarchical incremental label propagation method based on probability distribution, which can handle labels with progressive hierarchical relationships. Moreover, the probability distribution of each label of each node can be output to support subsequent tasks.

[0039] The content of the present application will be described in detail below with the help of specific embodiments.

[0040] Figure 1 is a flowchart of a hierarchical incremental label propagation method based on probability distribution according to an example embodiment. The hierarchical incremental label propagation method based on probability distribution 20 at least includes steps S102 to S108.

[0041] As Figure 1As shown, in S102, a relationship network is constructed through relationships among a plurality of users.

[0042] In the embodiments of the present application, the user can be a personal user or an enterprise user. The relationship among the users can be generated through user information. The user information can include basic information authorized by the user, such as business account information, terminal device identification information of the user, and regional information of the user. The user information can also include behavior information, such as page operation data of the user, business access time length of the user, and business access frequency of the user. The specific content of the user information can be determined according to the actual application scenario, which is not limited herein.

[0043] In S104, user labels are assigned to a first part of users in the relationship network according to a label policy. The target and dimension of the user labels to be assigned can be determined. The label policy is generated according to the target and the dimension. The user labels are assigned to the first part of users according to the label policy and historical data of a plurality of users in the relationship network.

[0044] In one embodiment, the historical data of a plurality of users can be obtained. The target data is extracted from the historical data according to the target in the label policy. The target data and the label policy are compared to assign user labels to the first part of users.

[0045] More specifically, the label interval can be determined according to the target and the dimension in the label policy. The target data and the label interval are compared. The historical users that meet the label interval are taken as the first part of users, and the labels are assigned to the first part of users according to the comparison result.

[0046] In one specific application, the dimension of the label can be set to 3, which can be mapped to label values 0, 1, and 2. The target of the user label can be whether the user is interested in a certain type of goods. Label value 0 can be set as not liking, label value 1 can be set as neutral, and label value 2 can be set as liking. The relevant data can be extracted from the historical shopping data of the user according to the target of the label. The label value of the user can be determined according to the purchase record of the user or the positive or negative evaluation result of the user. Of course, not all users in the plurality of sample users can extract the relevant data to determine the label. Only the first part of users that can determine the label are assigned the label. The users that are not assigned the label can be taken as the second part of users, and the labels of the second part of users can be determined according to subsequent steps.

[0047] In S106, a label dimension set of the second part of users in the relationship network is generated according to the user labels of the first part of users according to a hierarchical label propagation algorithm. The first part of users can be set as source nodes in the relationship network, for example; the label corresponding to the source node is propagated to the nodes corresponding to the second part of users in the relationship network according to the label propagation algorithm; when the variation parameter of the relationship network meets the threshold value, the propagation is stopped; and the label dimension set is generated according to the dimension and quantity of the label of the node corresponding to the second part of users.

[0048] The related content of "generating a label dimension set of the second part of users in the relationship network according to the probability distribution of the user labels of the first part of users according to a hierarchical label propagation algorithm" will be described in detail in the corresponding embodiments of Figure 3 .

[0049] In S108, the user label of the second part of users is determined according to the label dimension set. The multiple labels in the label set and the corresponding label quantity can be obtained; the probability corresponding to each label is calculated according to the multiple labels and the corresponding label quantity; and the user label of the second part of users is determined according to the probability corresponding to each label.

[0050] In one embodiment, the label level corresponding to each label can be extracted according to the label interval; the sum of the label probability of the current level and the label probability of the previous level is compared according to the label level; when the label probability of the current level is greater than the sum of the label probability of the previous level, the current label is taken as the user label of the second part of users.

[0051] Figure 2 is a schematic diagram of a hierarchical incremental label propagation method based on probability distribution according to another exemplary embodiment. The label determination method in the present application can be called the upper and maximum method, as Figure 2 indicated, each level of label in the label set is traversed, and if the probability of the current level is greater than the sum of the probabilities of all levels higher than the current level, the label of the current level is taken. The upper and maximum method is almost applicable to all progressive labels with pyramid-type distribution.

[0052] As Figure 2 indicated, assuming that the label of the current node is 2, if the total probability of the label greater than 2 is not less than the probability of the label being 2, the label of the current node should be determined as 3.

[0053] In an actual application scenario, multiple levels of labels can be set: label = 1 for normal users, label = 2 for users with slight abnormal behavior, label = 3 for users with more serious abnormal behavior, and label = 4 for users with definite bad behavior. Assuming that a user A has neighbor label statistics (10, 5, 3, 2), i.e., 10 normal users, 5 slightly abnormal users, 3 seriously abnormal users, and 2 users with definite bad behavior. If the maximum method is used, the label of A is 1, and if the upper and maximum method is used, the label of A is 3. The following are the iteration steps:

[0054] First round: assuming that the label of A is 1, i.e., A is a normal user, the probability is 10 / 20 = 50%; the probability that A is an abnormal user is (5+3+2) / 20 = 50%. Since the probability of a normal user is not greater than the probability of an abnormal user, A should be abnormal;

[0055] Second round: assuming that the label of A is 2, the probability is 5 / 20 = 25%; the probability that the label of A is greater than 2 is (2+3) / 20 = 25%; the probability of the original label is not greater than the sum of the probabilities of high-level labels, and A should take a high-level label;

[0056] Third round, assuming that the label of A is 3, the probability is 3 / 20 = 15%; the probability that the label of A is greater than 3 is 2 / 20 = 10%; the probability of the original label is greater than the sum of the probabilities of high-level labels, and the label of A takes 3;

[0057] The above steps can be understood as: the more associated users that are more abnormal than a certain user, the greater the risk of the user itself, and it is understandable that there are more normal associated users for a risk user, because in a normal scenario, the number of risk users is small.

[0058] According to the probability distribution-based hierarchical incremental label propagation method of the present application, a relationship network is constructed through the relationship between multiple users; a user label is assigned to a first part of users in the relationship network according to a label strategy; a label dimension set of a second part of users in the relationship network is generated through the user label of the first part of users according to a hierarchical label propagation algorithm; and a user label of the second part of users is determined according to the label dimension set, which can quickly and accurately determine the user label of a user without a label in actual application, meet the demand of calculation and analysis, and improve the calculation speed and reduce the occupation of content and computing resources compared with the label assignment method in the prior art.

[0059] It should be clearly understood that the present application describes how to form and use specific examples, but the principles of the present application are not limited to any details of these examples. On the contrary, based on the teachings of the disclosure of the present application, these principles can be applied to many other embodiments.

[0060] Figure 3 is a flow chart of a hierarchical incremental label propagation method based on probability distribution according to another exemplary embodiment. Figure 3 The flow 30 shown is a process of generating a label dimension set of a second part of users in a relationship network according to a hierarchical label propagation algorithm through a probability distribution of user labels of a first part of users. Figure 1 The detailed description of S106 in the flow shown, "generating a label dimension set of a second part of users in the relationship network according to a hierarchical label propagation algorithm through a probability distribution of user labels of a first part of users".

[0061] In the graph network, in the present application, nodes can be divided into three categories, source nodes, default nodes, and ordinary nodes.

[0062] Among them, the source node is the initial node with a label, the initial node; the default node is the initial node without a label, which can be converted into an ordinary node; the ordinary node is the default node which is given a label in the propagation process;

[0063] The node attribute can be designed as:

[0064]

[0065] For the multi-label problem, the present application considers that the label of a node is determined by the hidden variable of the node, that is, the label of the node is subject to a probability distribution. Therefore, the node attribute contains a probability distribution statistic array (prob). But prob records the distribution number of each label, not the direct probability, for the convenience of incremental updating, that is, prob can be regarded as a stock. In the present application, the use of distribution statistics prob as the description of the probability distribution of the hidden variable can directly output label and prob in the final output, or can be normalized to output probability prob and label label, and degree degree, through a propagation to obtain various analysis data, saving the calculation time and improving the calculation efficiency.

[0066] In the label propagation process, the final label (label) of each node can be recorded;

[0067] In the label propagation process, the state of the source node should always remain unchanged, so a label (isSource) can be added to the source node for separate processing in the propagation;

[0068] In the label propagation process, in order to realize dynamic propagation, the node can be set with a label (isActive) related to whether it is active;

[0069] In the label propagation process, in order to realize incremental propagation, the node can have a stock (prob) and an increment (diff) record.

[0070] Based on the above settings, as Figure 3As shown, in S302, the first part of users in the relationship network are set as source nodes.

[0071] In the relationship network, some nodes have labels that are marked as initial nodes, and the label is set to the corresponding known label; the initial node can be marked as a source node because the label propagation starts from them, and isSource=true is set.

[0072] In S304, according to the label propagation algorithm and the probability distribution, the label corresponding to the source node is propagated to the node corresponding to the second part of users in the relationship network in a hierarchical manner.

[0073] In one embodiment, the source node can be set as an active node, and the node corresponding to the second part of users can be set as an inactive node; the active node is extracted one by one, and the label array of the source node is passed to other nodes; other nodes generate their own corresponding temporary label according to the label array of the source node; when the temporary label and the last temporary label are different, the other nodes are set as active nodes.

[0074] The source node can be set as active to propagate in the first round of iteration;

[0075] The inventory array prob and the incremental array diff are set to 1 at the corresponding position of label, and the rest of the positions are set to zero.

[0076] In a specific embodiment, three labels can be included, mapped to 0, 1, and 2.

[0077] Then for a source node with label=1, its attributes can be set as:

[0078] (label=1, isSource=true, isActive=true, prob=[0,1,0], diff=[0,1,0]);

[0079] The attributes of the default node can be set as:

[0080] (label=-, isSource=false, isActive=false, prob=[0,0,0], diff=[0,0,0]).

[0081] The probability corresponding to each temporary label can be calculated according to the multiple temporary labels in the label data and the corresponding number of labels; the temporary label of this time for the second part of users is determined according to the probability corresponding to each temporary label; other nodes can update their own corresponding label array according to the label array of the source node.

[0082] For the source node: since its distribution is known, the prob inventory array does not need to be changed; in the incremental propagation mode, the source node is only propagated once, and then the active state is marked as inactive, and the increment is set to 0;

[0083] For the default node: the default node, i.e. the node that has never been active, is judged in the following way: the label value is still -1: all diff changes are accumulated to the prob statistics; the current label is obtained according to the prob statistics, and the label is set; the corresponding label position in diff is set to 1; and the active state is set.

[0084] For the ordinary node: all diff changes are accumulated to the prob statistics; the current label is obtained according to the prob statistics; if there is no change compared with the original label, the diff is set to 0, and the inactive state is set; otherwise, the new label corresponding position in diff is 1, and the old label corresponding position is -1, and the active state is set.

[0085] In S306, when the change quantity parameter of the relationship network meets the threshold value, the propagation is stopped. More specifically, the number of active nodes in the relationship network after each propagation is calculated; when the number of active nodes is less than the threshold value, the propagation is stopped.

[0086] After multiple iterations, the number of active nodes is less and less, or 0, or the number of active nodes before and after two rounds of iteration changes stably below a threshold value, at which time the iteration can be stopped. That is, the stop iteration condition can be set as: the number of active nodes, after two consecutive iterations, changes by no more than 5%.

[0087] In S308, the label dimension set is generated by the dimension and number of the label of the node corresponding to the second part of the user. The label label and the distribution statistics prob in the final node attribute are really useful information for subsequent tasks. That is, the output (label, prob). The sum of the distribution statistics is the degree of the current node. The output can also be arranged as a dimension set (label, prob / degree, degree).

[0088] According to the hierarchical incremental label propagation method based on probability distribution of the application, through the dynamic propagation and incremental propagation mode, the calculation resources, storage resources and bandwidth resources are greatly saved, so that the super large-scale relationship network containing data can be calculated on limited resources.

[0089] According to the hierarchical incremental label propagation method based on probability distribution of the application, by means of probability distribution, the hidden variable behind the node label is quantized as embedding, and the probability information of each label of the node is output, which can be used for subsequent downstream tasks.

[0090] According to the hierarchical incremental label propagation method based on probability distribution, the upper and maximum method is proposed, so that the label propagation algorithm can process labels with progressive hierarchical relationship.

[0091] Figure 4 is a flow chart of a hierarchical incremental label propagation method based on probability distribution according to another exemplary embodiment. Figure 4 The flow 40 shown is in actual operation, through specific array, variable, parameter programming to Figure 4 The implementation of the flow shown.

[0092] As Figure 4 As shown in S401, a relationship network is constructed.

[0093] In S402, the node attribute is initialized. For example, the default node attribute is initialized to be a non-active node; the source node attribute is initialized to be an active node.

[0094] In S403, the active node propagates the incremental array diff to all neighbor nodes, and this step can be incremental propagation.

[0095] In S404, the node receiving the message executes S505 if it is a source node, otherwise executes steps S406-S408.

[0096] In S405, the source node updates: the active state flag is set to non-active, and the increment diff is set to 0.

[0097] In S406, other nodes update: the increment is aggregated, and the increment and probability distribution statistics of the node are updated.

[0098] In S407, the new label of the node is determined according to the probability distribution statistics.

[0099] In S408, if the new and old labels are the same, the node is set to non-active, otherwise it is set to active.

[0100] In S409, the change of the number of active nodes in the relationship network is stopped within the threshold.

[0101] In S410, the attributes of each node are sorted and converted into output results.

[0102] In the hierarchical incremental label propagation method based on probability distribution, only the incremental array diff is propagated, so that the original information of each neighbor node can be saved, and the new probability distribution statistics can be calculated. At this time, the non-active nodes are silent, and only the neighbor nodes of the active nodes receive messages, which reduces the occupation of memory and computing resources.

[0103] It is worth mentioning that in the present application, in addition to the above and maximum method, the following methods can also be used:

[0104] Taking the label corresponding to the maximum value in the prob statistics;

[0105] Directly converting prob into a probability distribution, and randomly selecting a label according to the probability;

[0106] For progressive labels, convert prob into a probability distribution, take the expectation, and take the nearest label;

[0107] However, in actual application tests, the accuracy of the above methods is not as high as that of the above and maximum method in the present application.

[0108] In the present application, in the propagation process of the label, in addition to the direct addition method used in the present application, the following methods can also be used:

[0109] The adjacency matrix is normalized, and the corresponding weight exists on the edge as the weight of the diff transmitted by the corresponding edge, and weighted aggregation is performed;

[0110] According to other business rules, weighted aggregation is performed;

[0111] However, the incremental dynamic propagation method proposed in the present application uses the distribution statistics prob as the full amount, rather than directly using the probability, which can facilitate incremental accumulation calculation, simplify the problem difficulty, use diff as the incremental propagation, and only active nodes propagate externally, saving related resources.

[0112] Those skilled in the art can understand that all or part of the steps of the above embodiments are implemented as computer programs executed by a CPU. When the computer program is executed by the CPU, the above functions defined by the above method provided by the present application are executed. The program can be stored in a computer readable storage medium, which can be a read-only memory, a magnetic disk or an optical disk, etc.

[0113] In addition, it should be noted that the above figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present application, and are not for limiting purposes. It is easy to understand that the processes shown in the above figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be executed synchronously or asynchronously, for example, in multiple modules.

[0114] The following is a device embodiment of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

[0115] Figure 5is a block diagram of a probability distribution based hierarchical incremental label propagation device according to another exemplary embodiment. Figure 5 As shown in the figure, the probability distribution based hierarchical incremental label propagation device 50 comprises a network module 502, an assignment module 504, a propagation module 506 and a label module 508.

[0116] The network module 502 is configured to construct a relationship network through relationships among a plurality of users.

[0117] The assignment module 504 is configured to assign user labels to a first part of users in the relationship network according to a label policy.

[0118] The propagation module 506 is configured to generate a label dimension set of a second part of users in the relationship network through the user labels of the first part of users according to a hierarchical label propagation algorithm.

[0119] The label module 508 is configured to determine user labels for the second part of users according to the label dimension set.

[0120] According to the probability distribution based hierarchical incremental label propagation device of the present application, a relationship network is constructed through relationships among a plurality of users, user labels are assigned to a first part of users in the relationship network according to a label policy, and a label dimension set of a second part of users in the relationship network is generated through probability distributions of the user labels of the first part of users according to a hierarchical label propagation algorithm. In this way, user labels can be quickly and accurately determined for users without labels in actual applications, meeting the needs of calculation and analysis. Compared with the label assignment method in the prior art, the method in the present application improves the calculation speed and reduces the occupation of content and computing resources.

[0121] As shown in the figure, Figure 6 The present application provides an electronic device, comprising a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 complete mutual communication through the communication bus 640.

[0122] The memory 630 is configured to store a computer program.

[0123] The processor 610 is configured to execute the program stored in the memory 630, and implement the video expression action based data allocation permission adjustment method of any of the above embodiments.

[0124] The electronic device provided by the embodiment of the present application, the processor 610 obtains initial permissions and access information of the target data by executing the program stored in the memory 630; determines the video text content through the access information; establishes a real-time video link with the target, and displays the video text content according to the video connection to generate video data; identifies the expression action of the user in the video data to determine the corresponding permission adjustment coefficient; and adjusts the data permissions of the user according to the initial permissions and the permission adjustment coefficient.

[0125] The communication bus 640 mentioned above can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The communication bus 640 can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or only one type of bus.

[0126] The communication interface 620 is used for communication between the above-mentioned electronic device and other devices.

[0127] The memory 630 can include a random access memory (RAM), and can also include a non-volatile memory, such as at least one disk memory. Optionally, the memory 630 can also be at least one storage device located away from the aforementioned processor 610.

[0128] The processor 610 mentioned above can be a general-purpose processor 610, including a central processing unit (CPU), a network processor (NP), etc.; and can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0129] The embodiment of the present application provides a computer readable storage medium, and the computer readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors 1110 to implement the hierarchical incremental label propagation method based on a probability distribution of any one of the above-described embodiments. For example, a relationship network can be constructed through relationships among a plurality of users; user labels are assigned to a first part of users in the relationship network according to a label strategy; a label dimension set of a second part of users in the relationship network is generated through a probability distribution of the user labels of the first part of users according to a hierarchical label propagation algorithm; and user labels of the second part of users are determined according to the label dimension set.

[0130] In the above-described embodiments, the implementation can be achieved wholly or partially by software, hardware, firmware or any combination thereof. When implemented by software, the implementation can be achieved wholly or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the flow or function according to the embodiments of the present application is wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through a wired (for example, coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (for example, infrared, wireless, microwave, etc.) manner. The computer readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing one or more available media sets. The available medium can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), an optical medium (for example, DVD) or a semiconductor medium (for example, solid state disk (SSD)) and the like.

[0131] The exemplary embodiments of the present application are specifically shown and described above. It should be understood that the present application is not limited to the detailed structure, arrangement or implementation method described herein; on the contrary, the present application is intended to cover various modifications and equivalent arrangements included in the spirit and scope of the appended claims.

Claims

1. A hierarchical incremental label propagation method based on probability distribution, characterized in that, The application relates to a method for constructing a relationship network among a plurality of users, wherein the users include individual users and enterprise users, the relationship among the users is generated through user information, the user information includes business account information authorized by the users, terminal equipment identification information of the users, regional information of the users, page operation data, business access time length of the users and business access frequency of the users, a user label is assigned to a first part of users in the relationship network based on a label strategy, the user label includes whether the user is interested in a certain type of commodity, the label value of the user is determined according to purchase records of the user or positive and negative evaluation results of the user, the label value includes 0, 1 and 2, a label dimension set of a second part of users in the relationship network is generated through a probability distribution of the user label of the first part of users according to a hierarchical label propagation algorithm, a plurality of labels in the label dimension set and corresponding label quantities are obtained, the probability corresponding to each label is calculated according to the plurality of labels and the corresponding label quantities, the label level corresponding to each label is extracted according to a label interval, the sum of the label probability of the current level and the label probability of the previous level is compared according to the label level, and the current label is taken as the user label of the second part of users when the label probability of the current level is greater than the sum of the label probability of the previous level. The application relates to a method for constructing a relationship network among a plurality of users, wherein the users include individual users and enterprise users, the relationship among the users is generated through user information, the user information includes business account information authorized by the users, terminal equipment identification information of the users, regional information of the users, page operation data, business access time length of the users and business access frequency of the users, a user label is assigned to a first part of users in the relationship network based on a label strategy, the user label includes whether the user is interested in a certain type of commodity, the label value of the user is determined according to purchase records of the user or positive and negative evaluation results of the user, the label value includes 0, 1 and 2, a label dimension set of a second part of users in the relationship network is generated through a probability distribution of the user label of the first part of users according to a hierarchical label propagation algorithm, a plurality of labels in the label dimension set and corresponding label quantities are obtained, the probability corresponding to each label is calculated according to the plurality of labels and the corresponding label quantities, the label level corresponding to each label is extracted according to a label interval, the sum of the label probability of the current level and the label probability of the previous level is compared according to the label level, and the current label is taken as the user label of the second part of users when the label probability of the current level is greater than the sum of the label probability of the previous level. The application relates to a method for constructing a relationship network among a plurality of users, wherein the users include individual users and enterprise users, the relationship among the users is generated through user information, the user information includes business account information authorized by the users, terminal equipment identification information of the users, regional information of the users, page operation data, business access time length of the users and business access frequency of the users, a user label is assigned to a first part of users in the relationship network based on a label strategy, the user label includes whether the user is interested in a certain type of commodity, the label value of the user is determined according to purchase records of the user or positive and negative evaluation results of the user, the label value includes 0, 1 and 2, a label dimension set of a second part of users in the relationship network is generated through a probability distribution of the user label of the first part of users according to a hierarchical label propagation algorithm, a plurality of labels in the label dimension set and corresponding label quantities are obtained, the probability corresponding to each label is calculated according to the plurality of labels and the corresponding label quantities, the label level corresponding to each label is extracted according to a label interval, the sum of the label probability of the current level and the label probability of the previous level is compared according to the label level, and the current label is taken as the user label of the second part of users when the label probability of the current level is greater than the sum of the label probability of the previous level. ​ ​ ​ ​ ​ ​ 2. The method of claim 1, wherein, ​ ​ ​ ​ 3. The method of claim 2, wherein, ​ ​ ​ ​ 4. The method of claim 3, wherein, ​ ​ ​ ​ 5. The method of claim 1, wherein, ​ ​ ​ ​ ​ 6. The method of claim 5, wherein, According to the label propagation algorithm and the probability distribution, a label corresponding to the source node is propagated to nodes corresponding to the second part of users in the relationship network in a hierarchical manner, comprising: setting the source node as an active node and the nodes corresponding to the second part of users as non-active nodes; extracting the active nodes one by one and passing the label array of the source node to other nodes; generating a temporary label corresponding to the other nodes according to the probability distribution of the label array of the source node; setting the other nodes as active nodes when the temporary label is different from the last temporary label.

7. The method of claim 6, wherein, Generating a temporary label corresponding to the other nodes according to the probability distribution of the label array of the source node, comprising: updating the label array corresponding to the other nodes according to the label array of the source node; calculating the probability corresponding to each temporary label according to the multiple temporary labels and the corresponding label quantity in the label data; determining the temporary label for the second part of users according to the probability corresponding to each temporary label.

8. The method of claim 6, wherein, When the change quantity parameter of the relationship network meets the threshold value, stopping the propagation, comprising: calculating the change quantity of active nodes in the relationship network after each propagation; stopping the propagation when the change quantity of active nodes is less than the threshold value.

9. A hierarchical incremental tag propagation device based on probability distribution, characterized in that, Comprising: a network module for constructing a relationship network through the relationship between multiple users, wherein the users include individual users and enterprise users, the relationship between users is generated through user information, and the user information includes business account information authorized by the user, terminal device identification information of the user, regional information of the user, page operation data, business access time of the user, and business access frequency of the user; a distribution module for distributing user labels for a first part of users in the relationship network according to a label strategy, wherein the user labels include whether the user is interested in a certain type of goods, the label value of the user is determined according to the purchase record of the user or the positive or negative evaluation result of the user, and the label value includes 0, 1, and 2; a propagation module for generating a label dimension set of a second part of users in the relationship network through the probability distribution of the user labels of the first part of users according to a hierarchical label propagation algorithm; a label module for obtaining multiple labels and the corresponding label quantity in the label dimension set, calculating the probability corresponding to each label according to the multiple labels and the corresponding label quantity, extracting the label level corresponding to each label according to the label interval, comparing the sum of the label probability of the current level and the label probability of the previous level according to the label level, and taking the current label as the user label of the second part of users when the label probability of the current level is greater than the sum of the label probability of the previous level.

10. An electronic device, comprising: Comprising: one or more processors; a storage device for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1-8.

11. A computer readable medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the method of any one of claims 1-8.

Images