CN104615616A - Group recommendation method and system - Google Patents
Group recommendation method and system Download PDFInfo
- Publication number
- CN104615616A CN104615616A CN201410185426.9A CN201410185426A CN104615616A CN 104615616 A CN104615616 A CN 104615616A CN 201410185426 A CN201410185426 A CN 201410185426A CN 104615616 A CN104615616 A CN 104615616A
- Authority
- CN
- China
- Prior art keywords
- group
- module
- user
- vector
- location name
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a group recommendation method and system. The method includes the steps of obtaining user permanent point positions; obtaining position names corresponding to the user permanent point positions; according to the corresponding relation between the pre-stored position names and groups, obtaining the groups corresponding to the position names corresponding to the user permanent point positions to serve as candidate groups; ranking the candidate groups, and recommending the candidate groups to users. By means of the group recommendation method and system, the group related to the user permanent point positions can be recommended to the users, and the accuracy of group recommendation based on the positions can be improved.
Description
Technical field
The present invention relates to Internet technical field, particularly relate to a kind of group recommending method and system.
Background technology
Generally all there is a large amount of groups in social networks and JICQ, these groups can be made up of the user by identical hobby or same characteristic features, and the user in group can carry out the mutual of message with other member of group.In order to the group making user can find to meet self-demand in the group of magnanimity, need to recommend some groups to user.Traditional group recommending method is normally based on the personal information of user, the information such as such as hobby, age, occupation are recommended, such as group identical for hobby user can be recommended, and for example, the group relevant to occupation user can be recommended.
But, traditional group recommending method can not well for recommending some groups relevant with position, such as, group's (as xx area group), the group relevant with user job place (as xx industry park Major Leagues) etc. that user place of abode is relevant, these groups relevant with the position at user place are also that a lot of user wishes to know and to participate in usually, exactly this kind of group are recommended user and have great using value.
Summary of the invention
Based on this, be necessary for above-mentioned technical matters, a kind of group recommending method and the system that can recommend to reside with user to user the relevant group in a position are provided.
A kind of group recommending method, described method comprises:
Obtain user's resident some position;
Obtain the location name corresponding with described resident some position;
According to the corresponding relation of the location name prestored and group, obtain group alternatively group location name corresponding to corresponding with described resident some position;
Described candidate group is carried out sequence process and recommend user.
A kind of group commending system, described system comprises:
Position acquisition module, for obtaining user's resident some position;
Resident some location name acquisition module, for obtaining the location name corresponding with described resident some position;
Candidate's group determination module, for the corresponding relation according to the location name prestored and group, obtains the group alternatively group corresponding with the location name of described resident point;
Sort recommendations module, processes for described candidate group is carried out sequence and recommends user.
Above-mentioned group recommending method and system, a position can be resided according to user and obtain location name corresponding to resident some position, position-based title recommends to have the group of corresponding relation with location name to user, thus achieves and recommend to reside the relevant group in a position to user to user.The group relevant to position is determined by location name, can improve the accuracy that location-based group is recommended.
Accompanying drawing explanation
Fig. 1 is the applied environment figure of group recommending method in an embodiment;
Fig. 2 is the schematic flow sheet of group recommending method in an embodiment;
Fig. 3 is the schematic flow sheet excavating the group be correlated with in position in an embodiment;
Fig. 4 is the schematic flow sheet in Fig. 3, location name and group name being configured to respectively the semantic vector of fixed dimension;
Candidate group is carried out sequence to process and the schematic flow sheet recommending user by Fig. 5 in an embodiment;
Fig. 6 is the schematic flow sheet of the order models going out candidate group in Fig. 5 from training sample learning;
Fig. 7 is according to the schematic flow sheet that order models sorts to candidate group in Fig. 5;
Fig. 8 is the structured flowchart of group's commending system in an embodiment;
Fig. 9 is the structured flowchart of group's commending system in another embodiment;
Figure 10 is the structured flowchart of semantic vector constructing module in an embodiment;
Figure 11 is the structured flowchart of position acquisition module in an embodiment;
Figure 12 is the structured flowchart of sort recommendations module in an embodiment;
Figure 13 is the structured flowchart of order models study module in Figure 12;
Figure 14 is the structured flowchart of group's order module in Figure 12.
Embodiment
In order to make object of the present invention, technical scheme and advantage clearly understand, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, be not intended to limit the present invention.
The group recommending method that the embodiment of the present invention proposes can be applied in environment as shown in Figure 1.As shown in Figure 1, group's information bank stores group's information of magnanimity, and group's information comprises group name, group member list, group's introduction etc., and each group has unique group identification.The location name library storage location name of magnanimity, location name for representing certain position, such as " Shenzhen University ".The locating information of the user of location, LBS position library storage magnanimity, the positioning function in the terminal that these locating information use by user collects.Group excavates module can according to group's information bank and location name storehouse, the group relevant to position is excavated from group's information of magnanimity, such as to user job ground relevant group, the group etc. relevant with user settlement, then set up the corresponding relation of the location name group relevant to corresponding position.Customer location locating module obtains the resident some position of user by the locating information screening that LBS locates in storehouse, namely the position that often occurs of user, based on this position can recommend group excavate module the group relevant to position that excavate.Order module can carry out sequence process based on to determining to the candidate group of user's recommendation, and the group then order module sorted is outputted in various application by output interface.Be appreciated that group information bank, location name storehouse and storehouse, location, LBS position can be set in advance in various storage server, and can be arranged in various processing server for group excavation module, customer location locating module and the order module of data processing.
Shown in composition graphs 1, the group after the sequence that output interface exports can be used in multiple application, comprises group's recommendation, friend recommendation and network information push etc.Group recommends to refer to and the sort group of forward predetermined number of extracting directly can recommend user, friend recommendation refers to can be recommended user by the member in the forward multiple groups of sequence and add good friend, network information push refers to that the network information push multiple groups forward for sequence can be correlated with is to user, or the network information push member in multiple groups forward for sequence paid close attention to is to user.
As shown in Figure 2, in one embodiment, a kind of group recommending method provided, comprising:
Step 202, obtains user's resident some position.
Resident some position refers to the geographic position that user often occurs.Concrete, the geographic position at user place is obtained by position-based positioning service (Location Based Services, LBS).It can be then the geographic position that in user one day, maximum duration occurs that user resides a position, or the geographic position that user occurs in certain special time period.Such as, the geographic position that the position that in user one day, maximum duration occurs or user often occur in 10 to 7 these time periods of morning at night, then as user's resident some position.
Step 204, obtains the location name corresponding with resident some position.
Should there be corresponding location name in each geographic position, can be such as building name corresponding to geographic position or place title etc., such as " Shenzhen University ", " the life garden district that mattress is happy ", " Hill Park " etc.Each location name should have corresponding geographic position.Concrete, the different classes of location name corresponding with resident some position can be obtained according to the classification of location name, such as, need the location name obtaining place class, the location name corresponding with resident some position then can be obtained from the location name of place class, need the location name obtaining shelter class, then can obtain the location name corresponding with resident some position from the location name of shelter class.
Step 206, according to the corresponding relation of the location name prestored and group, obtains and resident alternatively group of group of putting corresponding to location name corresponding to position.
In the present embodiment, due to the group that the group recommending user is relevant to position, therefore can obtain from the group relevant to position of the magnanimity prestored and reside group corresponding to a position with user, these groups relevant to position can excavate in advance from the group of magnanimity.Such as, the group relevant to user's shelter can be " mattress happy life garden district group ", " Hohenschwangau community alliance group " etc., to user job relevant group can " Korean pine mansion cause group ", " foreign student's Chuangye Building group " etc.Usually, can comprise the information that some are relevant with position in the group information of the group be correlated with position, such as group name comprises location name, and group's brief introduction comprises location name etc.Therefore, the group meeting correspondence position title relevant to position, the corresponding relation between the group relevant to position and location name excavated can be set up in advance, and user resides some position correspondence location name, therefore group that can be corresponding according to the location name obtaining residing with user a position in this corresponding relation, alternatively group.
Step 208, carries out sequence by candidate group and processes and recommend user.
Concrete, can sort to candidate group according to some attribute informations of candidate group, these attribute informations include but not limited to that the geographic position that candidate group is corresponding and user reside the liveness etc. of distance between a position, candidate group.After sequence process, the candidate group that can extract the forward predetermined number of sequence is recommended.
In the present embodiment, a position can be resided according to user and obtain location name corresponding to resident some position, position-based title recommends to have the group of corresponding relation with location name to user, thus achieves and recommend to reside the relevant group in a position to user to user.The group relevant to position is determined by location name, can improve the accuracy that location-based group is recommended.
In one embodiment, the group relevant to position can be excavated in advance from the group of magnanimity.As shown in Figure 3, in one embodiment, the detailed process excavating the group relevant to position comprises:
Step 302, constructs the semantic vector of corresponding fixed dimension respectively to location name and group information.
Group information comprises the information such as group name, group's brief introduction.The semantic vector of fixed dimension that preferably can be corresponding to the group name structure in group information.Semantic vector refers to and represents location name and group information with vector, and this semantic vector can contain the context of co-text information of word.Preferably, this fixed dimension can be 200 dimensions.
In one embodiment, as shown in Figure 4, location name and group information are constructed respectively the step of the semantic vector of fixed dimension, comprising:
Step 402, carries out to location name and group information the set that participle obtains word.
Step 404, extracts the adjacent word of predetermined number before in the statement of word in default corpus, to the semantic vector of the adjacent word of the fixed dimension of the adjacent word structure correspondence extracted.
Step 406, is input to the semantic vector of adjacent word in predetermined probabilities model, by the parameter in the optimization aim determination predetermined probabilities model of probability model.
Preset in corpus and store a large amount of statements, data on the SNS social networks of specifying can be adopted as default corpus.In the present embodiment, adopt conditional probability to portray the context of co-text of word, the namely impact of word that occurred before being only subject to of the probability of each word, is designated as P (W
i| W
1, W
2.., W
i-1), such as sentence " vectorial cosine angle ", the probability of " angle " this word is subject to the impact of " cosine " and " vector ".Calculate for simplifying, can set the impact of n-1 the word that each word was only occurred above, conditional probability is P (W
i| W
i-n+
1...., W
i-1), wherein, the value that n can be set to 3 ~ 5, n is larger, then calculated amount is larger, but the quantity of information contained is larger, and namely discrimination is larger.Preferably n=3 can be set.
Preset in the statement in corpus, for each word wherein, it can be subject to the impact of n-1 word above, the word of n-1 being above input as each word of predetermined probabilities model, in this n-1 word, each word semantic vector represents, then the semantic vector of n-1 word forms a matrix [C (W
i-n+1) ...., C (W
i-1)], wherein C (W
i-1) be word W
i-1corresponding semantic vector.This n-1 semantic vector first place is connected, then forms the vector that (n-1) m ties up, be designated as x.Wherein, m is above-mentioned fixed dimension, is preferably 200.Then, be x modeling with a nonlinear hidden layer, i.e. tanh (Hx+d), wherein, d is bias term, and tanh is activation function.
Further, in one embodiment, the objective function Equation of probability model optimization is as follows:
y＝softmax(U·tanh(Hx+d)+Wx+b)
Wherein, softmax is activation function, U (| the matrix of V| × h, h is the number of plies of hidden layer) be the parameter of hidden layer to output layer; W (| the matrix of V| × (n-1) m) be directly to a matrix of a linear transformation of output layer from input layer.B is deviate.
The output layer of predetermined probabilities model is one | predicting the outcome of V| dimension, wherein V is the set of word.The i-th dimension y of this y that predicts the outcome
irepresent that next word is the probability of i, i.e. y
i=P (W
i| W
i-n+1...., W
i-1).Due to for the statement in default corpus, input layer and output layer are all determined.Therefore, the parameter in above-mentioned predetermined probabilities model can be trained according to the statement in corpus, i.e. U, W and b.
Step 408, by optimization object function, obtains the semantic vector of word.
After the parameter determining predetermined probabilities model, then by back-propagation algorithm optimization object function, thus obtain the semantic vector of each word after participle.Thus, semantic vector corresponding to location name be the semantic vector of the word comprised in location name and, the vector obtained after then doing normalized; The semantic vector that group information is corresponding be the semantic vector of the word comprised in group information and, then do the vector that normalized obtains.
Step 304, semantic vector corresponding for location name is carried out the training of one-class support vector machines sorter as training sample, semantic vector corresponding for group information input is completed in this one-class support vector machines sorter of training, has identified the group relevant to position.
In the present embodiment, adopt one-class support vector machines sorter (One class SVM), this sorter can calculate the hypersphere of positive sample in vector space, thus can identify with the similar test sample book of positive sample.In the present embodiment, using semantic vector corresponding for location name as positive sample.Be appreciated that and can determine that the semantic vector of which class location name is as positive sample according to the type of the group relevant to position wanting to recommend user.Such as, when the group needing to recommend is the group relevant to user's shelter, then can will represent that the semantic vector of the location name of shelter is as positive sample, when need the group recommended be with user job relevant group time, can represent that the semantic vector of the location name of office is as positive sample.Positive sample is input to one-class support vector machines sorter, the hypersphere in a vector space can be determined, this hypersphere by all Vertex cover of positive sample in the inside of ball.Semantic vector corresponding for group information is input in the one-class support vector machines sorter trained as test sample book, for a unknown test point (semantic vector that a group information is corresponding), if the inside of this ball determined at hypersphere, then think positive label.Accordingly, if in the outside of ball, then negative label is thought.The group that positive label is corresponding is then the group relevant to position.
Step 306, sets up the corresponding relation between the location name group relevant to position.
Concrete, the group relevant to position excavated must correspond to a location name, comprises this location name in the group information that this group is corresponding.In step 306, the group identification of this location name and corresponding group can be stored.
In conventional art, typically use 0/1 method for expressing and allow each word as an element of vector, the dimension of vector is then total word number of full text.When a certain bar text is expressed as vector, if the word corresponding to every one dimension of vector occurs in the text, then the value of this dimension is 1, otherwise is 0.This method for expressing due to the dimension of vector be total word number of full text, for the text of magnanimity, the dimension of vector is excessive thus cause calculated amount very large, thus the treatment effeciency of grievous injury sorter and performance.
In the present embodiment, by the method for expressing of the semantic vector of fixed dimension, predetermined probabilities model is utilized to be the expression that a vector row space found in each word, because semantic vector take into account the context of co-text of a word in statement, can calculated amount be reduced, the accuracy calculated can be ensured again.In addition, by one-class support vector machines sorter, semantic vector corresponding for the location name prestored is trained as training sample, semantic vector corresponding for group information input is completed the one-class support vector machines sorter of training, can directly obtain the group relevant to position, improve treatment effeciency.
In one embodiment, obtain the step that user resides a position, comprising: the customer location that receiving terminal reports; Cluster is carried out to the customer location reported, using the customer location in region the highest for concentration degree in cluster as user's resident some position.
Concrete, the customer location received is longitude and latitude, therefore customer location can form corresponding track on a plane space, Spatial Clustering can be adopted, as k-means clustering algorithm, cluster is carried out to the point on plane space, determine a region that concentration degree is the highest, the large I in this region pre-determines, thus using the customer location in this region as user's resident some position.Be appreciated that user's resident some position can for multiple.
Because user has different locating information at different time points, in the present embodiment, by carrying out cluster to the point of customer location on plane space, the position that user often occurs can be obtained, these positions are more may be user's shelter or office ground position, the recommendation of the group making this kind of position relevant is more accurate.
Further, the location name corresponding with resident some position can be obtained.Can according to store LBS position and building name corresponding relation database in, find and reside building name corresponding to a position with user.Because building name may have polytype, the building name of being such as correlated with according to shelter, with the relevant building name etc. in office space.The building name obtained can be input in the above-mentioned one-class support vector machines sorter trained, thus can obtain the building name of respective type, this building name is the location name corresponding with resident some position.
Further, owing to having prestored the corresponding relation of location name and group, can obtain the alternatively group of group corresponding to the location name corresponding with resident some position according to this corresponding relation, candidate group has been the group relevant to position.
In one embodiment, as shown in Figure 5, candidate group is carried out sequence and processes and the detailed process recommending user, comprising:
Step 502, using comprising the group of recommendation of positive sample and negative sample as training sample, goes out the order models of candidate group from this training sample learning.
In the present embodiment, positive sample is by the group of recommendation admitted, and negative sample is not by the group of recommendation admitted.Can be used for giving a mark to candidate group from the order models of the candidate group that training sample learns out, and this order models can make the candidate group that may be admitted by user sort more forward.
Step 504, sorts to candidate group according to order models.
Can determine according to order models the score value that candidate group is corresponding, thus can sort to candidate group according to the size of score value, the larger rank of score value is more forward.
Step 506, chooses the candidate group of the forward predetermined number of sequence and recommends user.
Concrete, in one embodiment, as shown in Figure 6, go out the step of the order models of candidate group from training sample learning, comprising:
Step 602, obtains the attributive character of training sample, according to attributive character structure attribute proper vector.
Attribute feature vector is made up of at least one attributive character, and attributive character is for characterizing the attribute of candidate group.Attributive character comprises following at least one: the liveness of group, the distance between group position and the resident some position of the corresponding user recommended, social topology information between group member with the corresponding user recommended.Wherein, the liveness of group is used for representing the active degree of group, can in group between member mutual size of message or this size of message is converted obtain, be certainly not limited thereto.Group position is by obtaining location name corresponding to group, thus the geographic position obtaining this location name corresponding obtains.Social topology information between group member with the corresponding user recommended includes but not limited in group member as in the quantity of good friend of this user, group member and size of message mutual between this user, mean distance etc. between group member and this user.
Step 604, constructs the mathematical model based on attribute feature vector.
In the present embodiment, this mathematical model is the attribute feature vector of training sample and the inner product of weight vector.Weight vector is the follow-up vector be used for the marking of candidate group, and this vector determines weights corresponding to different attributive character.
Step 606, according to the size order structure forecast collating sequence of mathematical model.
Prediction collating sequence formula is used for the sequence of the sequence of predicting candidate group, and it is hoped as far as possible close to the goal ordering sequence according to training sample training.
Step 608, obtains the goal ordering sequence that training sample is formed, according to the distance structure objective function between goal ordering sequence and prediction collating sequence.
In the present embodiment, group's record of the recommendation of a given m user, for each user i, the group whether admitting recommendation according to this user obtains above-mentioned training sample.Positive negative sample in training sample forms a goal ordering sequence, is designated as:
as basis recommends group to be sorted by the number of times that user admits, like this, the ratio negative sample of positive sample row is forward, wherein, and n
(i)represent the number of having recommended group.Group is recommended for each, corresponding attribute feature vector x can be obtained
j (i).Further, use mathematical model to calculate each ranking score having recommended group corresponding, this mathematical model is attribute feature vector x
j (i)with the inner product of weight vector, that is: f
w(x
j (i))=<w, x
j (i)>, wherein <, > represent inner product, and w is weight vector.Form prediction collating sequence according to the size of this ranking score, be designated as:
In order to make prediction collating sequence Z
(i)as far as possible close to goal ordering sequences y
(i), KL distance (cross entropy) between the sequence probability distribution of available targets collating sequence and prediction collating sequence, as loss function, portrays the gap of two sequences.Therefore, following formula can be adopted to obtain loss function:
Wherein, θ
k(j
1, j
2..., j
n) be k j before sequence
1, j
2..., j
kall sequences combination, P
sg () is a probability distribution function, be defined as:
Wherein, S
jfor the mark of a jth candidate feature vector, n
(i)it is candidate's number of user i.This probability function possesses characteristic: if positive sample row's is more forward, then this probable value is higher.Therefore, this probability function can well portray the effect of sequence.
Based on above-mentioned loss function, in whole training sample (such as, m user), the objective function of structure is then:
This target function type is a function asking weight vector w to be worth most.
Step 610, learns objective function, weight vector that is that obtain the above-mentioned distance making the user in training sample corresponding and that get minimum value.
Target function type due to machine learning is a function asking weight vector w to be worth most, and traditional gradient descent method (a kind of data ask the method be worth most) therefore can be utilized to obtain weight vector w to objective function.This weight vector can be used for determining the corresponding score value of candidate group.
Further, as shown in Figure 7, in one embodiment, the step sorted to candidate group according to order models, comprising:
Step 702, obtains the attribute feature vector that candidate group is corresponding.
Step 704, calculates the inner product of the weight vector attribute feature vector corresponding with candidate group, obtains the score value that candidate group is corresponding.
Because weight vector determines weights corresponding to different attributive character, weights are larger, then show that the sequence role of this attributive character to group is larger.The attribute feature vector that weight vector is corresponding with candidate group is multiplied, and the inner product obtained is score value corresponding to candidate group.Also namely: adopt formula f
w(x
j (i))=<w, x
j (i)> calculates, and wherein, w is weight vector, x
j (i)be a candidate group characteristic of correspondence attribute vector, score value corresponding to this candidate group is f
w(x
j (i)).
Step 706, the size according to score value sorts to corresponding candidate group.
Score value is larger, then show that user more may be interested in the candidate group of correspondence.The candidate group that can choose the forward predetermined number of sequence after sequence recommends user.
In the present embodiment, use the group of recommendation comprising positive negative sample as training sample, thus study can determine the weight vector of the weights of different attribute feature, sorts to candidate group according to this weight vector.The present embodiment has fully taken into account the impact of different attribute feature on the sequence of candidate group, makes the candidate's group user recommending user more possible interested, improves the accuracy that group is recommended.
The group recommending method that various embodiments of the present invention provide, can realize the group relevant with position to recommend user, is specially adapted to recommend some and user's shelter or user to handle official business relevant group.After obtaining the candidate group after sorting, except the forward candidate group of sequence can being recommended, be also applicable to other some application.Illustrate, can friend recommendation be realized: user except hope understand with oneself shelter or office except relevant group, also may wish to carry out deep communication with the member in these groups and exchange, therefore the member in candidate group forward for sequence can be recommended user and add good friend.Again such as, can also realize the propelling movement of the network information: the member in the candidate group sorting forward and user may be interested in a class network information, some network information push that therefore can be relevant by the position corresponding with candidate group are to group member and this user.
As shown in Figure 8, a kind of group commending system, this system comprises position acquisition module 802, resident some location name acquisition module 804, candidate's group determination module 806 and sort recommendations module 808, wherein:
Position acquisition module 802, for obtaining user's resident some position.
In the present embodiment, position acquisition module 802 can be used for the geographic position being obtained user place by position-based positioning service (LocationBased Services, LBS).
Resident some location name acquisition module 804, for obtaining the location name corresponding with resident some position.
Resident some location name acquisition module 804 may be used for obtaining the different classes of location name corresponding with resident some position according to the classification of location name, such as, resident some location name acquisition module 804 needs the location name obtaining place class, the location name corresponding with resident some position then can be obtained from the location name of place class, resident some location name acquisition module 804 needs the location name obtaining shelter class, then can be used for from the location name of shelter class, obtain the location name corresponding with resident some position.
Candidate's group determination module 806, for the corresponding relation according to the location name prestored and group, obtains the group alternatively group corresponding with the location name of resident point.
Sort recommendations module 808, processes for candidate group is carried out sequence and recommends user.
Sort recommendations module 808 can be used for sorting to candidate group according to some attribute informations of candidate group, and these attribute informations include but not limited to that the geographic position that candidate group is corresponding and user reside the liveness etc. of distance between a position, candidate group.After sequence process, the candidate group that sort recommendations module 808 can be used for extracting the forward predetermined number of sequence is recommended.
In one embodiment, as shown in Figure 9, this system also comprises group and excavates module 810, and this group excavates module 810 and comprises semantic vector constructing module 812, group identification module 814 and corresponding relation building module 816, wherein:
Semantic vector constructing module 812, for constructing the semantic vector of corresponding fixed dimension respectively to location name and group information.
In one embodiment, as shown in Figure 10, semantic vector constructing module 812 comprises word-dividing mode 812a, adjacent word extraction module 812b, parameter determination module 812c and semantic vector acquisition module 812d, wherein:
Word-dividing mode 812a, for carrying out to location name and group information the set that participle obtains word.
Adjacent word extraction module 812b, for extracting the adjacent word of predetermined number before in the statement of word in default corpus, to the semantic vector of the adjacent word of the fixed dimension of the adjacent word structure correspondence extracted.
Parameter determination module 812c, for being input in predetermined probabilities model by the semantic vector of adjacent word, by the parameter in the optimization aim determination predetermined probabilities model of probability model.
Semantic vector acquisition module 812d, for by optimization object function, obtains the semantic vector of word.
Group identification module 814, for semantic vector corresponding for location name to be carried out the training of one-class support vector machines sorter as training sample, semantic vector corresponding for group information input is completed in the one-class support vector machines sorter of training, has identified the group that position is relevant.
Positive sample is input to one-class support vector machines sorter by group identification module 814, can determine the hypersphere in a vector space, this hypersphere by all Vertex cover of positive sample in the inside of ball.Group identification module 814 is for being input in the one-class support vector machines sorter trained using semantic vector corresponding for group information as test sample book, for a unknown test point (semantic vector that a group information is corresponding), if the inside of this ball determined at hypersphere, then think positive label.Accordingly, if in the outside of ball, then negative label is thought.The group that positive label is corresponding is then the group relevant to position.
Corresponding relation building module 816, for setting up the corresponding relation between the location name group relevant to position.
Concrete, the group relevant to position that group's excavation module 810 is excavated must correspond to a location name, comprises this location name in the group information that this group is corresponding.Corresponding relation building module 816 can be used for storing the group identification of this location name and corresponding group.
In one embodiment, as shown in figure 11, position acquisition module 802 comprises position receiver module 802a and resident some position determination module 802b, wherein:
Position receiver module 802a, for the customer location that receiving terminal reports.
Resident some position determination module 802b, for carrying out cluster to the customer location reported, using the customer location in region the highest for concentration degree in cluster as user's resident some position.
In the present embodiment, the customer location that position receiver module 802a receives is longitude and latitude, therefore customer location can form corresponding track on a plane space, Spatial Clustering can be adopted, as k-means clustering algorithm, resident some position determination module 802b can be used for carrying out cluster to the point on plane space, determines a region that concentration degree is the highest, the large I in this region pre-determines, thus using the customer location in this region as user's resident some position.Be appreciated that user's resident some position can for multiple.
Further, resident some location name acquisition module 804 can be used for obtaining the location name corresponding with resident some position.Resident some location name acquisition module 804 according in the corresponding relation database of the LBS position stored and building name, can find and reside building name corresponding to a position with user.Resident some location name acquisition module 804 can be used for the building name obtained to be input in the above-mentioned one-class support vector machines sorter trained, thus the building name of respective type can be obtained, this building name is the location name corresponding with resident some position.
Further, owing to having prestored the corresponding relation of location name and group, candidate's group determination module 806 can obtain the alternatively group of group corresponding to the location name corresponding with resident some position according to this corresponding relation, and candidate group is the group relevant to position.
In one embodiment, as shown in figure 12, sort recommendations module 808 comprises: order models study module 808a, group order module 808b and group recommending module 808c, wherein:
Order models study module 808a, for using comprising the group of recommendation of positive sample and negative sample as training sample, goes out the order models of candidate group from training sample learning.
In the present embodiment, positive sample is by the group of recommendation admitted, and negative sample is not by the group of recommendation admitted.The order models of the candidate group that order models study module 808a learns out from training sample can be used for giving a mark to candidate group, and this order models can make the candidate group that may be admitted by user sort more forward.
Group order module 808b, for sorting to candidate group according to order models.
Group order module 808b can determine according to order models the score value that candidate group is corresponding, thus can sort to candidate group according to the size of score value, and the larger rank of score value is more forward.
Group recommending module 808c, for choosing the candidate group of the most forward predetermined number of sequence and recommending user.
Concrete, in one embodiment, as shown in figure 13, order models study module 808a comprises attribute feature vector constructing module 818a, mathematical model constructing module 828a, prediction collating sequence constructing module 838a, objective function module 848a and weight vector acquisition module 858a, wherein:
Attribute feature vector constructing module 818a, for obtaining the attributive character of training sample, according to attributive character structure attribute proper vector.
Mathematical model constructing module 828a, for constructing the mathematical model based on attribute feature vector, mathematical model is attribute feature vector and the inner product of corresponding weight vector.
Prediction collating sequence constructing module 838a, for the size order structure forecast collating sequence according to mathematical model.
Objective function module 848a, for obtaining the goal ordering sequence that training sample is formed, according to the distance structure objective function between goal ordering sequence and prediction collating sequence.
Attribute feature vector is made up of at least one attributive character, and attributive character is for characterizing the attribute of candidate group.Attributive character comprises following at least one: the liveness of group, the distance between group position and the resident some position of the corresponding user recommended, social topology information between group member with the corresponding user recommended.
In the present embodiment, group's record of the recommendation of a given m user, the group that each user i, objective function module 848a be can be used for whether admitting according to this user to recommendation obtains above-mentioned training sample.Positive negative sample in training sample forms a goal ordering sequence, is designated as:
as basis recommends group to be sorted by the number of times that user admits, like this, the ratio negative sample of positive sample row is forward, wherein, and n
(i)represent the number of having recommended group.Group is recommended for each, corresponding attribute feature vector x can be obtained
j (i).Further, objective function module 848a calculates each ranking score having recommended group corresponding by mathematical model, and this mathematical model is attribute feature vector x
j (i)with the inner product of weight vector, that is: f
w(x
j (i))=<w, x
j (i)>, wherein <, > represent inner product, and w is weight vector.Form prediction collating sequence according to the size of this ranking score, be designated as:
In order to make prediction collating sequence z
(i)as far as possible close to goal ordering sequences y
(i), objective function module 848a can be used for goal ordering sequence and prediction collating sequence sequence probability distribution between KL distance (cross entropy) as loss function, portray the gap of two sequences.Therefore, following formula can be adopted to obtain loss function:
Wherein, θ
k(j
1, j
2..., j
n) be k j before sequence
1, j
2..., j
kall sequences combination, P
sg () is a probability distribution function, be defined as:
Wherein, S
jfor the mark of a jth candidate feature vector, n
(i)it is candidate's number of user i.This probability function possesses characteristic: if positive sample row's is more forward, then this probable value is higher.Therefore, this probability function can well portray the effect of sequence.
Based on above-mentioned loss function, in whole training sample (such as, m user), the objective function that objective function module 848a constructs is then:
This target function type is a function asking weight vector w to be worth most.
Weight vector acquisition module 858a, for learning objective function, obtains weight vector that is that make distance corresponding to the user in training sample and that obtain minimum value.
Target function type due to machine learning is a function asking weight vector w to be worth most, and traditional gradient descent method (a kind of data ask the method be worth most) therefore can be utilized to obtain weight vector w to objective function.This weight vector can be used for determining the corresponding score value of candidate group.
Further, as shown in figure 14, group order module 808b comprises attribute feature vector acquisition module 818b, group point value computing module 828b and candidate group order module 838b, wherein:
Attribute feature vector acquisition module 818b, for obtaining attribute feature vector corresponding to candidate group.
Group point value computing module 828b, for calculating the inner product of the weight vector attribute feature vector corresponding with candidate group, obtains the score value that candidate group is corresponding.
Candidate group order module 838b, sorts to corresponding candidate group for the size according to score value.
Score value is larger, then show that user more may be interested in the candidate group of correspondence.The candidate group choosing the forward predetermined number of sequence after candidate group order module 838b is used in sequence recommends user.
One of ordinary skill in the art will appreciate that all or part of flow process realized in above-described embodiment method, that the hardware that can carry out instruction relevant by computer program has come, described program can be stored in a computer read/write memory medium, as in the embodiment of the present invention, this program can be stored in the storage medium of computer system, and performed by least one processor in this computer system, to realize the flow process of the embodiment comprised as above-mentioned each side method.Wherein, described storage medium can be magnetic disc, CD, read-only store-memory body (Read-Only Memory, ROM) or random store-memory body (Random Access Memory, RAM) etc.
The above embodiment only have expressed several embodiment of the present invention, and it describes comparatively concrete and detailed, but therefore can not be interpreted as the restriction to the scope of the claims of the present invention.It should be pointed out that for the person of ordinary skill of the art, without departing from the inventive concept of the premise, can also make some distortion and improvement, these all belong to protection scope of the present invention.Therefore, the protection domain of patent of the present invention should be as the criterion with claims.
Claims (16)
1. a group recommending method, described method comprises:
Obtain user's resident some position;
Obtain the location name corresponding with described resident some position;
According to the corresponding relation of the location name prestored and group, obtain group alternatively group location name corresponding to corresponding with described resident some position;
Described candidate group is carried out sequence process and recommend user.
2. method according to claim 1, is characterized in that, described method also comprises the step excavating the group be correlated with in position, specifically comprises:
Respectively described location name and group information are constructed to the semantic vector of corresponding fixed dimension;
Semantic vector corresponding for described location name is carried out the training of one-class support vector machines sorter as training sample, semantic vector corresponding for described group information input is completed in the described one-class support vector machines sorter of training, has identified the group relevant to position;
Set up the corresponding relation between the location name group relevant to described position.
3. method according to claim 2, is characterized in that, the described step described location name and group information being configured to respectively the semantic vector of fixed dimension, comprising:
The set that participle obtains word is carried out to location name and group information;
Extract the adjacent word of predetermined number before in the statement of institute's predicate in default corpus, to the semantic vector of the adjacent word of the fixed dimension of the adjacent word structure correspondence of described extraction;
The semantic vector of described adjacent word is input in predetermined probabilities model, by the parameter in the optimization aim determination predetermined probabilities model of described predetermined probabilities model;
By optimization object function, obtain the semantic vector of institute's predicate.
4. method according to claim 1, is characterized in that, described acquisition user resides the step of a position, comprising:
The customer location that receiving terminal reports;
Cluster is carried out to the described customer location reported, using the described customer location in region the highest for concentration degree in cluster as user's resident some position.
5. method according to claim 1, is characterized in that, described described candidate group is carried out sort process and recommend the step of user, comprising:
Using comprising the group of recommendation of positive sample and negative sample as training sample, go out the order models of candidate group from described training sample learning, wherein, described positive sample is by the group of recommendation admitted, and described negative sample is not by the group of recommendation admitted;
According to described order models, candidate group is sorted;
Choose the described candidate group of the most forward predetermined number of sequence and recommend user.
6. method according to claim 5, is characterized in that, the described step going out the order models of candidate group from described training sample learning, comprising:
Obtain the attributive character of described training sample, according to described attributive character structure attribute proper vector;
Construct the mathematical model based on described attribute feature vector, described mathematical model is the inner product of described attribute feature vector and weight vector;
According to the size order structure forecast collating sequence of described mathematical model;
Obtain the goal ordering sequence that training sample is formed, according to the distance structure objective function between described goal ordering sequence and prediction collating sequence;
Described objective function is learnt, weight vector that is that obtain the described distance making the user in described training sample corresponding and that obtain minimum value.
7. method according to claim 6, is characterized in that, the described step sorted to candidate group according to described order models, comprising:
Obtain the attribute feature vector that candidate group is corresponding;
Calculate the inner product of the described weight vector attribute feature vector corresponding with described candidate group, obtain the score value that described candidate group is corresponding;
Size according to described score value sorts to corresponding candidate group.
8. method according to claim 6, it is characterized in that, the attributive character comprised in described attribute feature vector comprises following at least one: the liveness of group, the distance between group position and the resident some position of the corresponding user recommended, social topology information between group member with the corresponding user recommended.
9. group's commending system, is characterized in that, described system comprises:
Position acquisition module, for obtaining user's resident some position;
Resident some location name acquisition module, for obtaining the location name corresponding with described resident some position;
Candidate's group determination module, for the corresponding relation according to the location name prestored and group, obtains the group alternatively group corresponding with the location name of described resident point;
Sort recommendations module, processes for described candidate group is carried out sequence and recommends user.
10. system according to claim 9, is characterized in that, described system also comprises group and excavates module, and described group excavates module and comprises:
Semantic vector constructing module, for constructing the semantic vector of corresponding fixed dimension respectively to described location name and group information;
Group identification module, for semantic vector corresponding for described location name to be carried out the training of one-class support vector machines sorter as training sample, semantic vector corresponding for described group information input is completed in the described one-class support vector machines sorter of training, has identified the group that position is relevant;
Corresponding relation building module, for setting up the corresponding relation between the described location name group relevant to described position.
11. systems according to claim 10, is characterized in that, described semantic vector constructing module comprises:
Word-dividing mode, for carrying out to location name and group information the set that participle obtains word;
Adjacent word extraction module, for extracting in the statement of predicate in default corpus before the adjacent word of predetermined number, to the semantic vector of the adjacent word of fixed dimension corresponding to the adjacent word structure of described extraction;
Parameter determination module, for being input in predetermined probabilities model by the semantic vector of described adjacent word, by the parameter in the optimization aim determination predetermined probabilities model of described predetermined probabilities model;
Semantic vector acquisition module, for by optimization object function, obtains the semantic vector of institute's predicate.
12. systems according to claim 9, is characterized in that, described position acquisition module comprises:
Position receiver module, for the customer location that receiving terminal reports;
Resident some position determination module, for carrying out cluster to the described customer position information reported, using the described customer location in region the highest for concentration degree in cluster as user's resident some position.
13. systems according to claim 9, is characterized in that, described sort recommendations module comprises:
Order models study module, for using comprising the group of recommendation of positive sample and negative sample as training sample, go out the order models of candidate group from described training sample learning, wherein, described positive sample is by the group of recommendation admitted, and described negative sample is not by the group of recommendation admitted;
Group's order module, for sorting to candidate group according to described order models;
Group's recommending module, for choosing the described candidate group of the most forward predetermined number of sequence and recommending user.
14. systems according to claim 13, is characterized in that, described order models study module comprises:
Attribute feature vector constructing module, for obtaining the attributive character of described training sample, according to described attributive character structure attribute proper vector;
Mathematical model constructing module, for constructing the mathematical model based on described attribute feature vector, described mathematical model is described attribute feature vector and the inner product of corresponding weight vector;
Prediction collating sequence constructing module, for the size order structure forecast collating sequence according to described mathematical model;
Objective function module, for obtaining the goal ordering sequence that training sample is formed, according to the distance structure objective function between described goal ordering sequence and prediction collating sequence;
Weight vector acquisition module, for learning described objective function, weight vector that is that obtain the described distance making the user in described training sample corresponding and that obtain minimum value.
15. systems according to claim 14, is characterized in that, described group order module comprises:
Attribute feature vector acquisition module, for obtaining attribute feature vector corresponding to candidate group;
Group point value computing module, for calculating the inner product of the described weight vector attribute feature vector corresponding with described candidate group, obtains the score value that described candidate group is corresponding;
Candidate group order module, sorts to corresponding candidate group for the size according to described score value.
16. systems according to claim 14, it is characterized in that, the attributive character comprised in described attribute feature vector comprises following at least one: the liveness of group, the distance between group position and the resident some position of the corresponding user recommended, social topology information between group member with the corresponding user recommended.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410185426.9A CN104615616B (en) | 2014-05-04 | 2014-05-04 | group recommendation method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410185426.9A CN104615616B (en) | 2014-05-04 | 2014-05-04 | group recommendation method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104615616A true CN104615616A (en) | 2015-05-13 |
| CN104615616B CN104615616B (en) | 2017-03-22 |
Family
ID=53150065
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410185426.9A Active CN104615616B (en) | 2014-05-04 | 2014-05-04 | group recommendation method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104615616B (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105337847A (en) * | 2015-11-17 | 2016-02-17 | 小米科技有限责任公司 | Instant messaging information processing method, device and equipment |
| CN106302085A (en) * | 2015-05-18 | 2017-01-04 | 腾讯科技（深圳）有限公司 | The recommendation method and system of instant messaging group |
| CN106385359A (en) * | 2016-10-06 | 2017-02-08 | 程在舒 | Group and friend automatic generation and photograph and video shared browsing method |
| CN106528860A (en) * | 2016-11-30 | 2017-03-22 | 华南师范大学 | Recommending method, device and system based on social network and big data analysis |
| CN106682035A (en) * | 2015-11-11 | 2017-05-17 | 中国移动通信集团公司 | Individualized learning recommendation method and device |
| CN106980703A (en) * | 2017-05-09 | 2017-07-25 | 北京三快在线科技有限公司 | For the method and device of group's search, electronic equipment, computer-readable medium |
| CN107391637A (en) * | 2017-07-10 | 2017-11-24 | 江苏省现代企业信息化应用支撑软件工程技术研发中心 | For possessing the group recommending method of geographical social information |
| CN108875993A (en) * | 2017-05-16 | 2018-11-23 | 清华大学 | Invite behavior prediction method and device |
| CN108876526A (en) * | 2018-06-06 | 2018-11-23 | 北京京东尚科信息技术有限公司 | Method of Commodity Recommendation, device and computer readable storage medium |
| CN109787784A (en) * | 2018-10-26 | 2019-05-21 | 深圳壹账通智能科技有限公司 | Group recommending method, device, storage medium and computer equipment |
| CN109949076A (en) * | 2019-02-26 | 2019-06-28 | 北京首钢自动化信息技术有限公司 | Establish method, information recommendation method and the device of hypersphere mapping model |
| CN110034941A (en) * | 2019-03-22 | 2019-07-19 | 北京众纳鑫海网络技术有限公司 | The method and system specific to the message group of equipment is added |
| CN110650428A (en) * | 2019-08-29 | 2020-01-03 | 惠州华阳通用电子有限公司 | Interaction method and device based on content and vehicle position |
| CN110889029A (en) * | 2018-08-17 | 2020-03-17 | 北京京东金融科技控股有限公司 | City target recommendation method and device |
| CN111130992A (en) * | 2019-11-22 | 2020-05-08 | 北京达佳互联信息技术有限公司 | Group recommendation method and device, electronic equipment and storage medium |
| CN112418402A (en) * | 2020-11-24 | 2021-02-26 | 百度在线网络技术（北京）有限公司 | Method for recommending object, neural network and training method thereof, and computing device |
| WO2021164619A1 (en) * | 2020-02-19 | 2021-08-26 | 北京达佳互联信息技术有限公司 | Group display method and device |
| CN113409158A (en) * | 2021-06-04 | 2021-09-17 | 上海明胜品智人工智能科技有限公司 | Channel object data processing and communication method, device, server and medium |
| CN113746648A (en) * | 2021-09-09 | 2021-12-03 | 武汉夜莺科技有限公司 | Information processing method, device and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120191777A1 (en) * | 2011-01-21 | 2012-07-26 | Qualcomm Incorporated | Contact relevance based on context |
| CN103096242A (en) * | 2011-11-03 | 2013-05-08 | 腾讯科技（深圳）有限公司 | User recommendation method and system based on location based services |
| CN103399900A (en) * | 2013-07-25 | 2013-11-20 | 北京京东尚科信息技术有限公司 | Image recommending method based on location service |
| CN103684994A (en) * | 2013-12-19 | 2014-03-26 | 宇龙计算机通信科技(深圳)有限公司 | Method and system for recommending friends for instant messaging users |
-
2014
- 2014-05-04 CN CN201410185426.9A patent/CN104615616B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120191777A1 (en) * | 2011-01-21 | 2012-07-26 | Qualcomm Incorporated | Contact relevance based on context |
| CN103096242A (en) * | 2011-11-03 | 2013-05-08 | 腾讯科技（深圳）有限公司 | User recommendation method and system based on location based services |
| CN103399900A (en) * | 2013-07-25 | 2013-11-20 | 北京京东尚科信息技术有限公司 | Image recommending method based on location service |
| CN103684994A (en) * | 2013-12-19 | 2014-03-26 | 宇龙计算机通信科技(深圳)有限公司 | Method and system for recommending friends for instant messaging users |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106302085A (en) * | 2015-05-18 | 2017-01-04 | 腾讯科技（深圳）有限公司 | The recommendation method and system of instant messaging group |
| CN106302085B (en) * | 2015-05-18 | 2020-08-25 | 腾讯科技（深圳）有限公司 | Recommendation method and system for instant messaging group |
| CN106682035A (en) * | 2015-11-11 | 2017-05-17 | 中国移动通信集团公司 | Individualized learning recommendation method and device |
| CN105337847A (en) * | 2015-11-17 | 2016-02-17 | 小米科技有限责任公司 | Instant messaging information processing method, device and equipment |
| CN106385359A (en) * | 2016-10-06 | 2017-02-08 | 程在舒 | Group and friend automatic generation and photograph and video shared browsing method |
| CN106528860A (en) * | 2016-11-30 | 2017-03-22 | 华南师范大学 | Recommending method, device and system based on social network and big data analysis |
| CN106980703A (en) * | 2017-05-09 | 2017-07-25 | 北京三快在线科技有限公司 | For the method and device of group's search, electronic equipment, computer-readable medium |
| CN106980703B (en) * | 2017-05-09 | 2020-06-30 | 北京三快在线科技有限公司 | Method and device for group search, electronic equipment and computer readable medium |
| CN108875993B (en) * | 2017-05-16 | 2022-05-10 | 清华大学 | Invitation behavior prediction method and device |
| CN108875993A (en) * | 2017-05-16 | 2018-11-23 | 清华大学 | Invite behavior prediction method and device |
| CN107391637A (en) * | 2017-07-10 | 2017-11-24 | 江苏省现代企业信息化应用支撑软件工程技术研发中心 | For possessing the group recommending method of geographical social information |
| CN108876526A (en) * | 2018-06-06 | 2018-11-23 | 北京京东尚科信息技术有限公司 | Method of Commodity Recommendation, device and computer readable storage medium |
| CN110889029A (en) * | 2018-08-17 | 2020-03-17 | 北京京东金融科技控股有限公司 | City target recommendation method and device |
| CN110889029B (en) * | 2018-08-17 | 2024-04-05 | 京东科技控股股份有限公司 | Urban target recommendation method and device |
| CN109787784A (en) * | 2018-10-26 | 2019-05-21 | 深圳壹账通智能科技有限公司 | Group recommending method, device, storage medium and computer equipment |
| CN109949076A (en) * | 2019-02-26 | 2019-06-28 | 北京首钢自动化信息技术有限公司 | Establish method, information recommendation method and the device of hypersphere mapping model |
| CN110034941A (en) * | 2019-03-22 | 2019-07-19 | 北京众纳鑫海网络技术有限公司 | The method and system specific to the message group of equipment is added |
| CN110650428A (en) * | 2019-08-29 | 2020-01-03 | 惠州华阳通用电子有限公司 | Interaction method and device based on content and vehicle position |
| CN110650428B (en) * | 2019-08-29 | 2021-06-29 | 惠州华阳通用电子有限公司 | Interaction method and device based on content and vehicle position |
| US11470032B2 (en) | 2019-11-22 | 2022-10-11 | Beijing Dajia Internet Information Technology Co., Ltd. | Method for recommending groups and related electronic device |
| CN111130992A (en) * | 2019-11-22 | 2020-05-08 | 北京达佳互联信息技术有限公司 | Group recommendation method and device, electronic equipment and storage medium |
| WO2021164619A1 (en) * | 2020-02-19 | 2021-08-26 | 北京达佳互联信息技术有限公司 | Group display method and device |
| CN112418402A (en) * | 2020-11-24 | 2021-02-26 | 百度在线网络技术（北京）有限公司 | Method for recommending object, neural network and training method thereof, and computing device |
| CN112418402B (en) * | 2020-11-24 | 2023-08-11 | 百度在线网络技术（北京）有限公司 | Method for recommending object, neural network, training method of neural network and computing device |
| CN113409158A (en) * | 2021-06-04 | 2021-09-17 | 上海明胜品智人工智能科技有限公司 | Channel object data processing and communication method, device, server and medium |
| CN113746648A (en) * | 2021-09-09 | 2021-12-03 | 武汉夜莺科技有限公司 | Information processing method, device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104615616B (en) | 2017-03-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104615616A (en) | Group recommendation method and system | |
| Yao et al. | Sensing spatial distribution of urban land use by integrating points-of-interest and Google Word2Vec model | |
| CN110399565A (en) | Based on when null cycle attention mechanism recurrent neural network point of interest recommended method | |
| Lee et al. | When twitter meets foursquare: tweet location prediction using foursquare | |
| Ke et al. | A CA-based land system change model: LANDSCAPE | |
| CN110765753B (en) | Document generation method, system, computer device and storage medium | |
| CN106326345A (en) | Friend relation mining method based on user behaviors in social network | |
| CN113360616A (en) | Automatic question-answering processing method, device, equipment and storage medium | |
| CN103703459A (en) | Method and system for text message normalization based on character transformation and unsupervised of web data | |
| Naboureh et al. | A review of land use/land cover change mapping in the China-Central Asia-West Asia economic corridor countries | |
| CN113127506B (en) | Target query statement construction method and device, storage medium and electronic device | |
| CN104834679A (en) | Representation and inquiry method of behavior track and device therefor | |
| CN111191099B (en) | User activity type identification method based on social media | |
| CN104731958A (en) | User-demand-oriented cloud manufacturing service recommendation method | |
| Li et al. | A regionalization method for clustering and partitioning based on trajectories from NLP perspective | |
| CN108062366A (en) | Public culture information recommendation system | |
| Charif et al. | A multi‐label cellular automata model for land change simulation | |
| CN111597321A (en) | Question answer prediction method and device, storage medium and electronic equipment | |
| Phillips et al. | Monitoring public engagement with nature using Google Trends | |
| CN114254615A (en) | Volume assembling method and device, electronic equipment and storage medium | |
| WO2020170020A1 (en) | Feedback mining with domain-specific modeling | |
| Imam et al. | Machine learning for characterizing growth in tourism employment in developing economies: an assessment of tourism employment in Sri Lanka | |
| Keramati et al. | Webpage clustering–taking the zero step: a case study of an Iranian website | |
| CN111177526B (en) | Network opinion leader identification method and device | |
| CN116089595A (en) | Data processing pushing method, device and medium based on scientific and technological achievements |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |