CN110163404A - A kind of diffusion of information prediction technique, device and server, storage medium - Google Patents

A kind of diffusion of information prediction technique, device and server, storage medium Download PDF

Info

Publication number
CN110163404A
CN110163404A CN201810606271.XA CN201810606271A CN110163404A CN 110163404 A CN110163404 A CN 110163404A CN 201810606271 A CN201810606271 A CN 201810606271A CN 110163404 A CN110163404 A CN 110163404A
Authority
CN
China
Prior art keywords
user
training
vector
message
target topic
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201810606271.XA
Other languages
Chinese (zh)
Inventor
高升
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tencent Technology Shenzhen Co Ltd
Beijing University of Posts and Telecommunications
Original Assignee
Tencent Technology Shenzhen Co Ltd
Beijing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tencent Technology Shenzhen Co Ltd, Beijing University of Posts and Telecommunications filed Critical Tencent Technology Shenzhen Co Ltd
Priority to CN201810606271.XA priority Critical patent/CN110163404A/en
Publication of CN110163404A publication Critical patent/CN110163404A/en
Pending legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/04Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/01Social networking

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Software Systems (AREA)
  • Tourism & Hospitality (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Artificial Intelligence (AREA)
  • Operations Research (AREA)
  • Game Theory and Decision Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Development Economics (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The embodiment of the invention discloses a kind of diffusion of information prediction techniques, device and server, storage medium, wherein method comprises determining that the target message under target topic, it obtains for indicating the first eigenvector of the first user and the second feature vector of second user in user's set under the target topic, obtain the theme feature vector for indicating the target topic, it calls and propagates estimation model to the first eigenvector, the second feature vector and the theme feature vector carry out that calculated result is calculated, wherein, the estimation model of propagating is pre-set for calculating the predictive information for propagating related news under the target topic in user's set between user, the DIFFUSION PREDICTION information of the target message is determined according to the calculated result, path when message is propagated in user's set can be achieved to predict.

Description

A kind of diffusion of information prediction technique, device and server, storage medium
Technical field
The present invention relates to machine learning techniques field more particularly to a kind of diffusion of information prediction technique, device and server, Storage medium.
Background technique
Currently universal with the continuous development of Smart-Its and smart machine, user from network or can pass through Other users obtain a large amount of message source, various message can also be transmitted to other users.Such as: user may be by oneself The news messages on periphery are sent to other users, it is also possible to forward the interested news messages in web browsing, and to disappear It ceases and is spread in user's circle where user, influenced so as to form certain public opinion.
Diffusion analysis for message such as news, advertisements is typically all specified at one according to the message at present Start to propagate in user's set (such as set that the correspondence user of each account is constituted in some social network-i i-platform), and passes through After a period of time, according to propagation condition of the message in user set, determine that the message collects in the user The spread conditions such as spread scope, propagation time in conjunction.And how to be diffused prediction to message in advance becomes current research Hot spot.
Summary of the invention
The embodiment of the invention provides a kind of diffusion of information prediction technique, device and server, storage mediums, it can be achieved that right The prediction of event information spread condition in user gathers.
On the one hand, the embodiment of the invention provides a kind of diffusion of information prediction techniques, comprising:
Target message under determining target topic in user's set;
It obtains for indicating that the first eigenvector of the first user and second is used in user's set under the target topic The second feature vector at family;
Obtain the theme feature vector for indicating the target topic;
Call propagate estimation model to the first eigenvector, the second feature vector and the theme feature to Amount carries out that calculated result is calculated, wherein the propagation estimation model is pre-set for calculating user's set The predictive information of related news under the target topic is propagated between middle user;
The DIFFUSION PREDICTION information of the target message is determined according to the calculated result.
On the other hand, the embodiment of the invention provides a kind of diffusion of information prediction meanss, comprising:
Determination unit, for determining the target message under target topic in user gathers;
Acquiring unit, for obtaining the fisrt feature for indicating the first user in user's set under the target topic The second feature vector of vector sum second user;
The acquiring unit is also used to obtain the theme feature vector for indicating the target topic;
Computing unit, for call propagate estimation model to the first eigenvector, the second feature vector and The theme feature vector carries out that calculated result is calculated, wherein the propagation estimation model is pre-set based on Calculate the predictive information for propagating related news under the target topic in user's set between user;
The determination unit is also used to determine the DIFFUSION PREDICTION information of the target message according to the calculated result.
In another aspect, the embodiment of the invention provides a kind of server, including processor and memory, the processor and The memory is connected with each other, wherein for storing computer program instructions, the processor is configured for the memory Described program instruction is executed, realizes above-mentioned diffusion of information prediction technique.
In another aspect, the embodiment of the invention provides a kind of computer readable storage medium, the computer-readable storage Media storage has computer program, and the computer program includes program instruction, and described program instructs when being executed by a processor The method for making the processor execute above-mentioned first aspect.
In embodiments of the present invention, it determines the target message under target topic, propagation estimation model can be called to acquisition For indicating that the theme of the feature vector of the first user and second user and the target topic is special in user's set under target topic Sign vector carries out that calculated result is calculated, then the DIFFUSION PREDICTION letter of the target message can be determined according to the calculated result Breath, to realize the prediction of the spread condition to target message in user's set.
Detailed description of the invention
Technical solution in order to illustrate the embodiments of the present invention more clearly, below will be to needed in embodiment description Attached drawing is briefly described, it should be apparent that, drawings in the following description are some embodiments of the invention, general for this field For logical technical staff, without creative efforts, it is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of schematic flow diagram of diffusion of information prediction technique provided in an embodiment of the present invention;
Fig. 2 is the schematic flow diagram of the acquisition methods of the user characteristics vector of the embodiment of the present invention;
Fig. 3 is the schematic flow diagram of the training method of the associated vector of the embodiment of the present invention;
Fig. 4 a is a kind of diffusion of information network diagram provided in an embodiment of the present invention;
Fig. 4 b is that a diffusion of information path of diffusion of information network as shown in fig. 4 a provided in an embodiment of the present invention shows It is intended to;
Fig. 4 c is another diffusion of information path of diffusion of information network as shown in fig. 4 a provided in an embodiment of the present invention Schematic diagram;
Fig. 5 is a kind of diffusion of information path schematic diagram provided in an embodiment of the present invention;
Fig. 6 is a kind of schematic block diagram of diffusion of information prediction meanss provided in an embodiment of the present invention;
Fig. 7 is a kind of schematic block diagram of server provided in an embodiment of the present invention.
Specific embodiment
In embodiments of the present invention, the various types of message such as news messages, advertisement information needs propagated, Can these message start propagate before or propagate complete before, or even propagate complete after, as needed to message Spread condition is predicted, in embodiments of the present invention, when the expansion for needing a certain target message under the target topic to propagation When scattered situation is predicted, can by under target topic indicate user set in the feature of the first user and second user to The theme feature vector of amount and the target topic calls and propagates estimation model and be calculated one for being predicted Calculated result, and then determine the target message in the user gathers between two users according to the calculated result Predictive information, the predictive information are mainly used to predict diffusion feelings of the target message between the first user and second user Condition, such as the predictive information may include indicating that the target message is broadcast to the letter such as probability of second user from the first user Breath.Each in user's set with can be used as the first user or second user per family, user's collection available in this way The a large amount of predictive information about target message propagation condition closed are based on a large amount of predictive information, can predict to obtain the mesh Situations such as marking spread condition of the message in user set, such as can determining the number scale of diffusion, diffusion distance.
In one embodiment, indicated under target topic the first user first eigenvector and second user second Feature vector presets to obtain, and the feature vector of the first user under the target topic for uniquely indicating described the One user, and the feature vector of second user is then used to uniquely indicate the second user under the target topic.Difference is main Topic can predefine unique theme feature vector, for example, pre-set for uniquely indicate the theme feature of current political news to Amount, pre-sets the theme feature vector for uniquely indicating sports news.In one embodiment, the theme of target topic The user characteristics vector of feature vector, different user under target topic is pre- to first pass through model to train generation.
In one embodiment, which specifically can be the propagation estimation model.Propagation estimation model is used to Probability value or distance value (Euclidean distance value) are calculated based on vector, and user's set includes the first eigenvector, the The feature vector of all users including two feature vectors, the theme feature vector of the target topic, is all based on the biography Estimation model is broadcast come what is trained, specifically can be the N message based on target topic in user set after propagation, Based on known propagation path, to the initial vector of the initial vector sum target topic of each user in user's set It is trained, finally obtains the theme feature of feature vector and the target topic of each user under the target topic Vector.
Estimate model come to message to based on feature vector and theme feature vector, propagation first in embodiments of the present invention Propagation carry out prediction be illustrated, as shown in Figure 1, providing a kind of schematic flow diagram of diffusion of information prediction technique, the party Method can complete to target message user set in spread condition predict, first in S101, determine target topic Under target message, such as some news messages under current political news, and obtaining for indicating user under the target topic The second feature vector of the first eigenvector of first user and second user in set.First eigenvector and second feature to Amount can distinguish the different user under same target topic.It include a large amount of user in user's set, these users can be The user of specified group, such as specified student group user, specified worker's group of subscribers or certain being manually specified It may check a bit, the user of communication target message.First user and second user can be any two in user's set and use Family.The user characteristics vector of each user equal existence anduniquess under the target topic in user's set.The institute of first user It states first eigenvector and can refer to the related of subsequent embodiment to the specific acquisition modes of the second feature vector of second user and retouch It states.
After obtaining the first eigenvector and second feature vector, obtain in S102 for indicating the target The theme feature vector of theme.In one embodiment, the theme feature vector of different target theme is different, but same target master It is identical for inscribing the theme feature vector of lower different target message, for example, the theme feature vector e under current political news theme1 It indicates, the theme feature vector under entertainment news theme can use e2It indicates.Equally, the theme feature vector of different themes is specific Acquisition modes can refer to the description of subsequent embodiment.
It should be noted that the sequencing that the execution of step S101 and step S102 is not necessarily to, can be performed simultaneously step Rapid S101 and step S102, can also first carry out step S101, then execute step S102, or first carry out step S102, then execute Step S101.
The embodiment of the present invention is in the case where getting the target topic for indicating that first of the first user in user's set is special Vector is levied, after the theme feature vector of the second feature vector sum target topic of second user, in S103, calls and propagates estimation Model carries out the first eigenvector, the second feature vector and the theme feature vector calculating knot is calculated Fruit, wherein propagation estimation model is pre-set to propagate the mesh between user for calculating in user's set The predictive information of related news under theme is marked, which refers to any message under the target topic.
In one embodiment, the calculated result (calculated result i.e. the predictive information for propagating estimation model) It can be the information an of probability or Euclidean distance etc, and propagation estimation model then can be accordingly for based on probability It calculates or the objective function apart from calculating is come the model constructed in advance, propagation estimation model is for above-mentioned first referred to The model that feature vector, second feature vector and theme feature vector are calculated.In one embodiment, it is believed that first User is sender user, and second user is recipient user, and the first eigenvector can use Y1It indicates, described the Two feature vectors use Y2It indicates, the theme feature vector of the target topic uses e1Indicate, propagate estimation model be according to Predictive information is calculated in the objective function for calculating distance, such as the objective function may is that Loss=| | Y1+e1-Y2 ||2, a distance value can be calculated based on the propagation estimation model that the objective function constructs, the distance value can be made For predictive information, to estimate whether the target message can travel to second user from the first user.In one embodiment, such as Fruit is smaller (being less than preset distance threshold) based on above-mentioned LOSS function calculated distance value, it may be considered that first User is very likely broadcast to second user, and prediction result can be determined as to the first user to be broadcast to second for target message User, and if LOSS function calculated distance value is larger (being greater than preset distance threshold), it may be considered that first It is smaller that user is broadcast to a possibility that second user, will not even be broadcast to second user, then prediction result can be determined Target message will not be broadcast to second user for the first user.
Each in user set with can be used as sender user i.e. the first user, and each user per family It can also be used as recipient user i.e. second user.By above-mentioned step, any two user in user's set is calculated Between predictive information.The DIFFUSION PREDICTION information for determining the target message according to the calculated result in S104, that is, exist Global analysis is carried out based on a large amount of predictive information in S104, is determined in user gathers, the target message can be at which It is propagated between user, the DIFFUSION PREDICTIONs information such as range size (such as quantity of the user of propagation) of propagation.
In embodiments of the present invention, determine the target message under target topic, can call propagation estimation model to one or Possible communication target message is predicted between each user in user's set of the multiple desired dispensings of person, it may be determined that about target topic Target message DIFFUSION PREDICTION information, grasp accurate the analysis of public opinion data, be conducive to the accurate dispensing of certain message, no Message can be launched to the user group that may not propagate certain message.
Fig. 2 is referred to again, is first eigenvector and the second spy of second user of the first user of the embodiment of the present invention The schematic flow diagram of the acquisition methods of vector is levied, the acquisition methods can correspond to the S101 in above-described embodiment, firstly, In S201, obtain for indicate the second basis of the first foundation vector sum second user of the first user in user set to Amount.In embodiments of the present invention, be different from the above-mentioned first eigenvector referred to and second feature vector, the first foundation to Amount is for indicating first user in user gathers, and second basis vector is for indicating institute in user gathers Second user is stated, the different representation methods that different user uses in user set ands, which is realized, gathers the user The differentiation of middle user, wherein whether second basis vector of first foundation vector sum is propagated with the first user and second user Theme corresponding to some message is unrelated.
Obtain for indicate user set in the first user first foundation vector sum second user second basis to After amount, the mapping function under target topic can be obtained, the mapping function includes sender's mapping function and reception in S202 Square mapping function, and the first foundation vector is mapped according to the mapping function in S203 to obtain first spy Vector is levied, and second basis vector is mapped to obtain the second feature vector.That is, if described first User is as sender user, and the second user is as recipient user, i.e., according to described sender mapping function to described First foundation vector is mapped to obtain the first eigenvector, and according to recipient's mapping function to second base Plinth vector is mapped to obtain second feature vector.It in one embodiment, can be by the first of sender user (the first user) Basis vector is denoted as u1, the second basis vector of recipient user's (second user) is denoted as u2, and can will be under the target topic Sender's mapping function is denoted as w1, recipient's mapping function is denoted as w2, then according to described sender mapping function w1By described first Basis vector u1It is mapped to obtain the first eigenvector w of first user1u1, and according to recipient's mapping function w2By the second basis vector u2It is mapped to obtain the second feature vector w of the second user2u2
In embodiments of the present invention, by obtaining the base for indicating the first user and second user in user's set Mapping function under target topic described in plinth vector sum, can map to obtain the feature of first user and the second user to Amount, and the user in user set can be traversed according to the mode and obtain in user's set each user in some theme Under user characteristics vector, due to the difference that is indicated according to the basis vector of user of feature vector of user set, and Mapping function can flexibly be changed based on different themes, so that the method is bigger about the scalability of different themes.
Fig. 3 is referred to again, is the schematic flow diagram of the training method of the associated vector of the embodiment of the present invention.Based on the instruction Practicing method can train to obtain in user's set under the associated vector of multiple users, the theme feature vector of various themes and theme The mapping function being related to.
When being trained to associated vector, it is necessary first to determine that training supervision message obtains the target in S301 Diffusion path of the training message in communication process under theme, is determined according to the diffusion path for indicating the training Message first user from user set propagates to the training supervision message of the second user.Based on known diffusion road Diameter come to training exercise supervision, the user A in the diffusion path is bound to be broadcast to user B, that is to say, that user A and use The probability propagated between the B of family or apart from the vector for coming training user A, user B and theme feature as training supervision message.
The trained message is the training message under the target topic, and the trained message can be gathered in user In user between be diffused, spread result related data be used as train supervision message.In one embodiment, may be used Training message is propagated in above-mentioned user set, is determined in user's set further according to the searching mark of the trained message The user identifier of the trained message has been propagated, and has been established according to the user identifier and the trained transmission of news direction The diffusion path of the trained message, wherein the searching mark includes the keyword and/or label of the trained message, because This, as long as any one user's click has been read about the content of the trained message or in the instruction in user set Practice to make comments under message and content or as comment content issue or the training disappears using the trained message Breath has been sent to other users as forwarding content, all can keyword based on the trained message and/or label determine institute Whether the user stated in user's set has received the training message, or whether has sent the training message, and then can determine that The user identifier of the trained message is propagated in user's set and the user gathers the direction for propagating the trained message, Obtain a plurality of propagation path.
It in one embodiment, can be by the keyword and/or Hashtag of the event information (training message) under different themes Partly or entirely the click reading content of user, comment content and forwarding content carry out in (Hash label) and the user set Matching similarly can extract multiple out to extract the diffusion of information path in user's set about the trained message The diffusion of information path in the diffusion of information path of event information, the multiple event information may make up different event information (still Belonging to target topic) diffusion of information network in user group (user's set) is as shown in fig. 4 a two trained message pair The diffusion of information network answered includes that (its dispersal direction is respectively with diagram for two diffusion paths in the diffusion of information network The instruction of the arrow of middle solid line and dotted line), wherein two diffusion paths are as shown in figures 4 b and 4 c.
Fig. 4 a is that the event X that will belong under target topic and event Y is sent to user group M (the user group M includes User A, B, C, D, E) in be trained, by will be in the keyword and/or label of event X and event Y and the user group M Click reading information, comment information and/or the forwarding information of user matches, and can extract event as shown in Figure 4 b respectively The diffusion path of the diffusion path of X and event Y as illustrated in fig. 4 c, the diffusion of the diffusion path of the event X and the event Y Path constitutes diffusion of information network of the X and Y event information in user group M.So as to based on spreading known to above-mentioned two Path is special to train the theme of theme corresponding to the user characteristics vector sum event X of user A, B, C, D, E in determining user group M Levy the theme feature vector of theme corresponding to vector, event Y.
It obtains training supervision message based on diffusion path, that is to say, that show in diffusion path adjacent in diffusion path Two users between had propagated corresponding trained message, as shown in Figure 4 b, if using a certain Euclidean distance value as distance Show to have propagated corresponding training message between two users when threshold value, then, the training supervision letter between user A and user B Breath is the value less than or equal to distance threshold, and similarly, the training supervision message between user B and user C is to be less than or wait Training supervision message between the value of distance threshold, user C and user E is the value less than or equal to distance threshold.Similarly Training supervision message between the corresponding two vector users of available Fig. 4 c.
Further in S302, the theme training vector under the target topic is obtained, the theme training vector is made E ' can be used for the initial vector of the target topic1It indicates.In one embodiment, the theme training vector under target topic can also To be obtained by the skipgram method in word2vec, or obtained by way of generating at random, theme training to Amount needs to carry out subsequent training as training data, to obtain can be ultimately utilized in the predictive information being calculated between two users Theme feature vector.
After determining the theme training vector of the target topic, in S303, call the propagation estimation model to institute State first training vector, the second user second instruction under the target topic of first user under the target topic The theme training vector for practicing target topic described in vector sum carries out that trained calculated result is calculated.In one embodiment, institute Before stating S303, can include determining that the first user the first training vector and second user the second training vector the step of, It may particularly include: obtaining the training mapping function under the target topic;It will be described first according to the trained mapping function The first training basis vector that user generates is mapped to obtain first training vector, and by the second of the second user Training basis vector is mapped to obtain second training vector.First training basis vector is intended to indicate that described first uses The initial vector at family, the second training basis vector are intended to indicate that the initial vector of the second user.
The trained mapping function includes sender's training mapping function and recipient's training mapping function, if described the One user is as sender user, and the second user is as recipient user, then first training vector is according to What sender's mapping function and the first training basis vector generated, second training vector is reflected according to the recipient Penetrating function and described second trains basis vector to generate.
It is (every based on the path point in the diffusion path after the diffusion path for determining the training message under target topic One path point indicates a user) carry out random walk, can formation sequence data, then be directed to the data of the serializing Using skipgram method (a kind of generation in word2vec (a kind of vector field homoemorphism type being understood that for generating computer) The method for the vector that computer is understood that), obtain each user in user set distributed low-dimensional vector indicate to get The first training basis vector and the second training basis vector before to training, in order to further obtain the depth of user and event Semantic feature indicates (i.e. the first training vector of the first user and the second training vector of second user).In one embodiment In, the first training basis vector can be denoted as u '1, the second training basis vector is denoted as u '2, mapped for trained sender Function is denoted as w '1, w ' is denoted as trained recipient's mapping function2, then the first training vector generated by mapping is denoted as w′1u′1, the second training vector is denoted as w '2u′2
In one embodiment, based on training message, if its propagation path is as shown in figure 5, can be obtained the propagation road The first training foundation characteristic vector u ' of neighboring user A and B on diameter1With the second training foundation characteristic vector u '2, and in event master Inscribe on relevant latent sheaf space based on training mapping function (including sender training mapping function w '1It is mapped with recipient's training Function w '2) mapped, obtaining new user characteristics indicates (the first i.e. above-mentioned training vector w '1u′1With the second training vector W '2U '2).In one embodiment, after the diffusion path for determining the trained message, the first training base of first user Second training basis vector of plinth vector sum second user can also be randomly generated.First training basis vector and the second training Basis vector is trained as the trained values of user characteristics, with obtain can be ultimately utilized in be calculated it is pre- between two users The first eigenvector and second feature vector of measurement information.
The trained calculated result can be based on above-mentioned first training vector, the second training vector and theme training to A trained probability being calculated or training Euclidean distance are measured, and propagation estimation model is then accordingly for based on upper The model for stating the training objective function of trained probability calculation or training distance calculating to construct in advance.In one embodiment, such as Above-mentioned, the expression of the first training vector is for w '1u′1, the second training vector is expressed as w '2u′2, theme training vector is expressed as e '1, Described propagate estimates that model may is that Loss=according to the training objective function for calculating distance | | w '1u′1+e′1-w′2u′2| |2
In one embodiment, when the training calculated result and the training for calling the propagation estimation model to be calculated When supervision message is consistent or closely similar, then illustrate the first training basis vector, the second training basis vector, transmission at this time Fang Xunlian mapping function, recipient's training mapping function and theme training vector can effectively obtain correct calculated result, That is the message of meeting communication target theme between two users, these vectors are effectively, can be directed to again other including described in The diffusion path of first user and second user carrys out the above-mentioned calculating of further progress, so that u ' under each path1、w′1、u′2、 w′2And e '1It is all satisfied calculated result and the training supervision consistent condition of message, and then from u '1、w′1、u′2、w′2And e '1Instruction Get final u1、w1、u2、w2And e1
If the trained calculated result and the training supervision message are inconsistent, need in S304, according to described Training supervision message and the trained calculated result, to the training of first training vector, the second training vector and the theme Vector is updated, to obtain optimal vector.In one embodiment, when the training supervision message and the training calculate knot When fruit is inconsistent, can to it is above-mentioned first training basis vector, second training basis vector, training mapping function and theme training to One or more in amount carries out vector update, calls and propagates estimation model based on updated u '1、w′1、u′2、w′2And e′1It is calculated, so that consistent with training supervision message according to the calculated result that estimation model is calculated is propagated.
In one embodiment, when first user is as sender user, the second user is recipient's use Family includes: then to be updated to obtain sender to described sender training mapping function to the update of first training vector Mapping function, and the first training basis vector is updated to obtain above-mentioned first foundation vector;To second training It includes: to be updated to obtain recipient's mapping function to recipient training mapping function that vector, which is updated, and to described Second training basis vector is updated to obtain above-mentioned second basis vector.
In embodiments of the present invention, by obtaining diffusion road of the training message under the target topic in communication process Diameter, it may be determined that out for indicating that the trained message first user from user set propagates to the instruction of the second user Practice supervision message, then can obtain the theme training vector under the target topic, and calls propagation estimation model to described first Training vector, the second training vector and the theme training vector carry out that trained calculated result is calculated, when the training meter Calculate result and when the inconsistent supervision message, it is trained to first training vector, the second training vector and the theme to Amount is updated, so that it is consistent with the training supervision message using the result that updated vector is calculated, it realizes To the continuous adjustment for propagating estimation model, so that the accuracy for being diffused prediction to the target message can be improved.
The embodiment of the invention also provides a kind of diffusion of information prediction meanss, the diffusion of information prediction meanss are for executing The unit of aforementioned described in any item methods.It specifically, is that a kind of diffusion of information provided in an embodiment of the present invention is pre- referring to Fig. 6 Survey the schematic block diagram of device.The diffusion of information prediction meanss of the present embodiment comprise determining that unit 601, acquiring unit 602 and meter Calculate unit 603.In embodiments of the present invention, the diffusion of information prediction meanss can be set needs to carry out message dissemination some In the server of prediction or some dedicated pre- measurement equipments.
Determination unit 601, for determining the target message under target topic;Acquiring unit 602, for obtaining in institute Stating indicates the first eigenvector of the first user and the second feature vector of second user in user's set under target topic;It is described Acquiring unit 602 is also used to obtain the theme feature vector for indicating the target topic;Computing unit 603, for calling It propagates estimation model the first eigenvector, the second feature vector and the theme feature vector calculate To calculated result, wherein propagation estimation model is pre-set to pass between user for calculating in user's set Broadcast the predictive information of related news under the target topic;The determination unit 601 is also used to be determined according to the calculated result The DIFFUSION PREDICTION information of the target message.
In one embodiment, the acquiring unit 602, for obtaining for indicating user under the target topic The first eigenvector of first user and when the second feature vector of second user in set, for obtaining for indicating the use Second basis vector of the first foundation vector sum second user of the first user in the set of family;Obtain reflecting under the target topic Penetrate function;Mapped to obtain the first eigenvector to the first foundation vector according to the mapping function, and by institute The second basis vector is stated to be mapped to obtain the second feature vector.
In one embodiment, the mapping function includes sender's mapping function and recipient's mapping function;When described First user is as sender user, and when the second user is as recipient user, the acquiring unit 602 is reflected according to described It penetrates function the first foundation vector is mapped to obtain the first eigenvector, and second basis vector is carried out Mapping obtains the concrete mode of the second feature vector are as follows: according to described sender mapping function to the first foundation vector It is mapped to obtain the first eigenvector, and second basis vector is reflected according to recipient's mapping function It penetrates to obtain the second feature vector.
In one embodiment, the diffusion of information prediction meanss further include: updating unit 604, the acquiring unit 602, it is also used to obtain diffusion path of the training message in communication process under the target topic, the determination unit 601 It is determined according to the diffusion path for indicating that it is described that the trained message first user from user set propagates to The training supervision message of second user;The acquiring unit 602, be also used to obtain under the target topic theme training to Amount;The computing unit 603 is also used to call the propagation estimation model to first user under the target topic The theme of the second training vector and the target topic of first training vector, the second user under the target topic is instructed Practice vector to carry out that trained calculated result is calculated;Updating unit 604, for according to the training supervision message and the training Calculated result is updated first training vector, the second training vector and the theme training vector.
In one embodiment, the diffusion of information device further include: map unit 605, the acquiring unit 602, also For obtaining the training mapping function under the target topic;Map unit 605, for that will be according to the trained mapping function The first training basis vector that first user generates is mapped to obtain first training vector, and will be described second The second training basis vector that user generates is mapped to obtain second training vector.
In one embodiment, the trained mapping function includes sender's training mapping letter under the target topic Several and recipient training mapping function;When first user is as sender user, first training vector is basis What described sender training mapping function and the first training basis vector generated, then the updating unit 604 is specifically used for: Be updated to obtain sender's mapping function to described sender training mapping function, and to the first training basis vector into Row, which updates, obtains first foundation vector;When the second user is as recipient user, then second training vector is root It is generated according to recipient training mapping function and the second training basis vector, the updating unit 604 is specifically used for: Be updated to obtain recipient's mapping function to recipient training mapping function, and to the second training basis vector into Row, which updates, obtains the second basis vector.
In one embodiment, the acquiring unit 602 is being passed in the training message for obtaining under the target topic When diffusion path during broadcasting, for propagating the training message under the target topic in user set;According to institute The searching mark for stating trained message determines the user identifier that the trained message is propagated in user's set, and according to institute State the diffusion path that user identifier establishes the trained message.
In one embodiment, the searching mark includes the keyword and/or label of the trained message, the determination Unit 601 for the searching mark according to the trained message, is being determined to propagate the trained message in user's set User identifier when, for the corresponding behaviour of user in gathering the keyword of the trained message and/or label with the user Retrieval matching is carried out as content;It determines to propagate the trained message in user's set based on matched matching result is retrieved User identifier, and determine the direction of propagation;The operation content that the user generates includes: the click reading content of user, comment Inside perhaps forward at least one arrived in content.
In embodiments of the present invention, determination unit 601 determines the target message under the target topic, acquiring unit 602 The first eigenvector and second user for indicating the first user in user's set under the target topic can be obtained Second feature vector, and obtain the theme feature vector for indicating the target topic, then computing unit 603 can call biography It broadcasts estimation model theme feature vector described in the first eigenvector, the second feature vector sum be calculated It calculates as a result, the DIFFUSION PREDICTION information for making the determination unit 601 that can determine the target message according to the calculated result, real The prediction to the diffusion path of target message is showed.
It is a kind of schematic block diagram of server provided in an embodiment of the present invention referring to Fig. 7.In the present embodiment as shown in the figure Server may include: power supply, shell, the structures such as various required interfaces, such as network interface, user interface etc. Deng.The server further include: one or more processors 701 and storage device 702.Above-mentioned processor 701 and storage device 702 are connected, and in one embodiment, can be connected by bus 703 between processor 701 and storage device 702.
The server may include user interface, the user interface may include some physical buttons or touch by The interface module of the compositions such as key can receive the operation of user, which can also can prompt the user with including some The structures such as the display screen of the information such as the working condition of server.
The storage device 702 may include volatile memory (volatile memory), such as random access memory (random-access memory, RAM);Storage device 702 also may include nonvolatile memory (non-volatile Memory), such as flash memory (flash memory), solid state hard disk (solid-state drive, SSD) etc.;Storage device 702 can also include the combination of the memory of mentioned kind.
The processor 701 can be central processing unit (central processing unit, CPU).The processor 701 can further include hardware chip.Above-mentioned hardware chip can be specific integrated circuit (application- Specific integrated circuit, ASIC), programmable logic device (programmable logic device, PLD) etc..The PLD can be field programmable gate array (field-programmable gate array, FPGA), lead to With array logic (generic array logic, GAL) etc..The combination of the processor 701 or above structure.
In embodiments of the present invention, for the memory 702 for storing computer program, the computer program includes journey Sequence instruction, processor 701 is used to execute the program instruction of the storage of memory 702, for realizing the respective party in above-described embodiment Method step.
In one embodiment, the processor 701 is configured to call described program instruction, for determining under target topic Target message;Obtain the first eigenvector and second for indicating the first user in user's set under the target topic The second feature vector of user;Obtain the theme feature vector for indicating the target topic;It calls and propagates estimation model pair The first eigenvector, the second feature vector and the theme feature vector carry out that calculated result is calculated, In, propagation estimation model is pre-set to propagate the target topic between user for calculating in user's set The predictive information of lower related news;The DIFFUSION PREDICTION information of the target message is determined according to the calculated result.
In one embodiment, the processor 701 is for obtaining for indicating user's collection under the target topic The first eigenvector of first user and when the second feature vector of second user in conjunction, for obtaining for indicating the user Second basis vector of the first foundation vector sum second user of the first user in set;Obtain the mapping under the target topic Function;The first foundation vector is mapped according to the mapping function to obtain the first eigenvector, and will be described Second basis vector is mapped to obtain the second feature vector.
In one embodiment, the mapping function includes sender's mapping function and recipient's mapping function, when described First user is as sender user, and when the second user is as recipient user, the processor 701 is also used to according to institute It states sender's mapping function the first foundation vector is mapped to obtain the first eigenvector, and according to the reception Square mapping function maps second basis vector to obtain the second feature vector.
In one embodiment, the processor 701 is also used to obtain the training message under the target topic and is propagating Diffusion path in the process is determined according to the diffusion path for indicating the trained message the from user set One user propagates to the training supervision message of the second user;Call the propagation estimation model to first user in institute State the second training vector and the mesh of the first training vector, the second user under the target topic under target topic The theme training vector of mark theme carries out that trained calculated result is calculated;According to the training supervision message and the training meter It calculates as a result, being updated to first training vector, the second training vector and the theme training vector.
In one embodiment, the processor 701 is also used to obtain the training mapping function under the target topic;Root It is mapped the first training basis vector generated for first user to obtain described first according to the trained mapping function Training vector, and by the second training basis vector generated for the second user mapped to obtain second training to Amount.
In one embodiment, the trained mapping function includes sender's training mapping letter under the target topic Several and recipient training mapping function, when first user is as sender user, first training vector is basis What described sender training mapping function and the first training basis vector generated, the processor 701 is also used to described Sender's training mapping function is updated to obtain sender's mapping function, and is updated to the first training basis vector Obtain first foundation vector;When first user is as sender user, the second user is as recipient user, then Second training vector is generated according to recipient training mapping function and the second training basis vector, described Processor 701 is also used to be updated to obtain to the recipient training mapping function recipient's mapping function, and to described the Two training basis vectors are updated to obtain the second basis vector.
In one embodiment, the processor 701 is being propagated in the training message for obtaining under the target topic When diffusion path in the process, for propagating the training message under the target topic in user set;According to described The searching mark of training message determines the user identifier that the trained message is propagated in user's set, and according to described User identifier establishes the diffusion path of the trained message.
In one embodiment, the searching mark includes the keyword and/or label of the trained message, the processing Device 701 for the searching mark according to the trained message, is being determined to propagate the trained message in user's set When user identifier, for the keyword of the trained message and/or label operation content corresponding with the user to be examined Rope matching;The user identifier that the trained message is propagated in user's set is determined based on matched matching result is retrieved, And determine the direction of propagation;The operation content that the user generates includes: the click reading content of user, in comment perhaps in forwarding At least one of arrived in appearance.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with Relevant hardware is instructed to complete by computer program, the program can be stored in a computer-readable storage medium In, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be magnetic Dish, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random Access Memory, RAM) etc..
Above disclosed is only section Example of the invention, cannot limit the right of the present invention with this certainly Range, those skilled in the art can understand all or part of the processes for realizing the above embodiment, and according to right of the present invention Equivalent variations made by it is required that, still belongs to the scope covered by the invention.

Claims (11)

1. a kind of diffusion of information prediction technique characterized by comprising
Determine the target message under target topic;
Obtain the first eigenvector and second user for indicating first user in user's set under the target topic Second feature vector;
Obtain the theme feature vector for indicating the target topic;
Call propagate estimation model to the first eigenvector, the second feature vector and the theme feature vector into Calculated result is calculated in row, wherein propagations estimation model be it is pre-set for calculate the user gather in use The predictive information of related news under the target topic is propagated between family;
The DIFFUSION PREDICTION information of the target message is determined according to the calculated result.
2. the method according to claim 1, wherein the acquisition is for indicating user under the target topic The second feature vector of the first eigenvector of first user and second user in set, comprising:
Obtain the second basis vector for indicating the first foundation vector sum second user of the first user in user's set;
Obtain the mapping function under the target topic;
Mapped to obtain the first eigenvector to the first foundation vector according to the mapping function, and by described Two basis vectors are mapped to obtain the second feature vector.
3. according to the method described in claim 2, it is characterized in that, the mapping function includes sender's mapping function and reception Square mapping function;
It is described to be reflected according to when the second user is as recipient user when first user is as sender user It penetrates function the first foundation vector is mapped to obtain the first eigenvector, and second basis vector is carried out Mapping obtains the second feature vector and includes:
The first foundation vector is mapped according to described sender mapping function to obtain the first eigenvector, and root Second basis vector is mapped according to recipient's mapping function to obtain the second feature vector.
4. the method according to claim 1, wherein the method also includes:
Diffusion path of the training message in communication process under the target topic is obtained, is determined according to the diffusion path For indicating that the trained message first user from user set propagates to the training supervision message of the second user;
Obtain the theme training vector under the target topic;
Call the estimation model of propagating to first training vector of first user under the target topic, described second The theme training vector of second training vector and the target topic of the user under the target topic carries out that instruction is calculated Practice calculated result;
According to the training supervision message and trained calculated result, to first training vector, second training to Amount and the theme training vector are updated.
5. according to the method described in claim 4, it is characterized in that, the method also includes:
Obtain the training mapping function under the target topic;
It is mapped the first training basis vector generated for first user to obtain institute according to the trained mapping function The first training vector is stated, and is mapped the second training basis vector generated for the second user to obtain second instruction Practice vector.
6. according to the method described in claim 4, it is characterized in that, the trained mapping function is included under the target topic Sender training mapping function and recipient training mapping function;
When first user is as sender user, first training vector is according to described sender training mapping letter What the several and described first training basis vector generated, being updated to first training vector includes: to instruct to described sender Practice mapping function to be updated to obtain sender's mapping function, and the first training basis vector is updated to obtain first Basis vector;
When the second user is as recipient user, second training vector is according to recipient training mapping letter What the several and described second training basis vector generated, being updated to second training vector includes: to instruct to the recipient Practice mapping function to be updated to obtain recipient's mapping function, and the second training basis vector is updated to obtain second Basis vector.
7. according to the method described in claim 4, it is characterized in that, the training message obtained under the target topic is passing Diffusion path during broadcasting, comprising:
The training message under the target topic is propagated in user set;
According to the searching mark of the trained message, user's mark that the trained message is propagated in user's set is determined Know, and establishes the diffusion path of the trained message according to the user identifier.
8. the method according to the description of claim 7 is characterized in that the searching mark includes the keyword of the trained message And/or label, the searching mark according to the trained message, it determines to propagate the trained message in user's set User identifier, comprising:
The operation content that the keyword of the trained message and/or label are generated with user in user set is retrieved Matching;
The user identifier that the trained message is propagated in user's set is determined based on the matched matching result of retrieval, And determine the direction of propagation;
The operation content that the user generates include: the click reading content of user, perhaps forward in content in comment arrive to One item missing.
9. a kind of diffusion of information prediction meanss characterized by comprising
Determination unit, for determining the target message under target topic;
Acquiring unit, for obtaining the first eigenvector for indicating the first user in user's set under the target topic With the second feature vector of second user;
The acquiring unit is also used to obtain the theme feature vector for indicating the target topic;
Computing unit propagates estimation model to the first eigenvector, the second feature vector and described for calling Theme feature vector carries out that calculated result is calculated, wherein the propagation estimation model is pre-set for calculating institute State the predictive information for propagating related news under the target topic in user's set between user;
The determination unit is also used to determine the DIFFUSION PREDICTION information of the target message according to the calculated result.
10. a kind of server, which is characterized in that including processor and memory, the processor is mutually interconnected with the memory It connects, wherein the memory is configured for execution described program and refers to for storing computer program instructions, the processor It enables, realizes the method according to claim 1.
11. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage has computer journey Sequence, the computer program include program instruction, and described program instruction executes the processor such as The described in any item methods of claim 1-8.
CN201810606271.XA 2018-06-12 2018-06-12 A kind of diffusion of information prediction technique, device and server, storage medium Pending CN110163404A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810606271.XA CN110163404A (en) 2018-06-12 2018-06-12 A kind of diffusion of information prediction technique, device and server, storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810606271.XA CN110163404A (en) 2018-06-12 2018-06-12 A kind of diffusion of information prediction technique, device and server, storage medium

Publications (1)

Publication Number Publication Date
CN110163404A true CN110163404A (en) 2019-08-23

Family

ID=67644982

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810606271.XA Pending CN110163404A (en) 2018-06-12 2018-06-12 A kind of diffusion of information prediction technique, device and server, storage medium

Country Status (1)

Country Link
CN (1) CN110163404A (en)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102831234A (en) * 2012-08-31 2012-12-19 北京邮电大学 Personalized news recommendation device and method based on news content and theme feature
CN104933622A (en) * 2015-03-12 2015-09-23 中国科学院计算技术研究所 Microblog popularity degree prediction method based on user and microblog theme and microblog popularity degree prediction system based on user and microblog theme
CN106372437A (en) * 2016-09-07 2017-02-01 北京邮电大学 Information diffusion prediction method and device
CN106600071A (en) * 2016-12-20 2017-04-26 西南石油大学 Topic-based microblog forwarding prediction method and system
CN107330562A (en) * 2017-07-03 2017-11-07 扬州大学 Information dissemination method based on individual consumer's feature
CN107908789A (en) * 2017-12-12 2018-04-13 北京百度网讯科技有限公司 Method and apparatus for generating information
CN107944635A (en) * 2017-12-13 2018-04-20 福州大学 A kind of information propagation forecast model and method for merging the topic factor

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102831234A (en) * 2012-08-31 2012-12-19 北京邮电大学 Personalized news recommendation device and method based on news content and theme feature
CN104933622A (en) * 2015-03-12 2015-09-23 中国科学院计算技术研究所 Microblog popularity degree prediction method based on user and microblog theme and microblog popularity degree prediction system based on user and microblog theme
CN106372437A (en) * 2016-09-07 2017-02-01 北京邮电大学 Information diffusion prediction method and device
CN106600071A (en) * 2016-12-20 2017-04-26 西南石油大学 Topic-based microblog forwarding prediction method and system
CN107330562A (en) * 2017-07-03 2017-11-07 扬州大学 Information dissemination method based on individual consumer's feature
CN107908789A (en) * 2017-12-12 2018-04-13 北京百度网讯科技有限公司 Method and apparatus for generating information
CN107944635A (en) * 2017-12-13 2018-04-20 福州大学 A kind of information propagation forecast model and method for merging the topic factor

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
GAO SHENG 等: "A Novel Embedding Method for Information Diffusion Prediction in Social Network Big Data", IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, vol. 13, pages 2097 - 2105, XP011658216, DOI: 10.1109/TII.2017.2684160 *
李志清;: "基于LDA主题特征的微博转发预测", 情报杂志, no. 09, pages 158 - 162 *

Similar Documents

Publication Publication Date Title
US9252961B2 (en) Identify experts and influencers in a social network
CN103842983B (en) For building and send the method and system of information flow
CN103918001B (en) Comprehension effects of communications propagated through social networking system
JP6145576B2 (en) Large page recommendation in online social networks
Yin et al. Spatio-temporal recommendation in social media
CN103716338B (en) A kind of information-pushing method and device
CN106605418A (en) Power management of mobile clients using location-based services
JP2014194796A (en) Methods and systems for determining geographic user profile to determine suitability of targeted content messages based on the profile
US20150242447A1 (en) Identifying effective crowdsource contributors and high quality contributions
CN107925701A (en) Notified for the event of application
Mayfield et al. Predictive risk mapping of an environmentally-driven infectious disease using spatial Bayesian networks: A case study of leptospirosis in Fiji
JP2011530772A (en) Identifying objects in an image
KR20110073282A (en) Consolidating input messages for social activity summarization
US20120226521A1 (en) Utilize Experts and Influencers in a Social Network
US10463975B2 (en) Automatic fantasy sports data analysis method and apparatus
CN104158748B (en) A kind of topological detecting method towards system for cloud computing
CN107545451A (en) A kind of advertisement sending method and device
CN107545444A (en) A kind of card data recommendation method and device
Barnwal et al. Enhancing reliability of vehicular participatory sensing network: A Bayesian approach
CN104111981A (en) Method and device used for providing post messages
Patten et al. Biogeographical boundaries and Monmonier’s algorithm: a case study in the northern Neotropics
WO2017206566A1 (en) Message propagation method, and server and computer storage medium
CN105879393A (en) Webpage game service server and webpage game event reminding method and system
CN107545453A (en) A kind of information distribution method and device
CN110297967A (en) Point of interest determines method, apparatus, equipment and computer readable storage medium

Legal Events

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