CN104601438A - Friend recommendation method and device - Google Patents
Friend recommendation method and device Download PDFInfo
- Publication number
- CN104601438A CN104601438A CN201410180918.9A CN201410180918A CN104601438A CN 104601438 A CN104601438 A CN 104601438A CN 201410180918 A CN201410180918 A CN 201410180918A CN 104601438 A CN104601438 A CN 104601438A
- Authority
- CN
- China
- Prior art keywords
- user
- candidate
- characteristic vector
- commending friends
- friends
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention applies to the field of internet, and provides friend recommendation method and device. The method comprises the steps of acquiring data information of a user; determining candidate recommended friends of the user and the feature vectors of the candidate recommended friends according to the data information of the user, wherein the feature vectors include quantifying information of personal property of the user, and the quantifying information of social networking services topological structure; calculating the scores corresponding to the feature vectors according to the pre-trained model so as to obtain the score sequence of the friends of the user; recommending friends to the user according to the calculated score sequence of the friends of the user. According to the method and device, the personal property information of the user and the topological network structure information are combined, and thus comprehensive recommended parameter data are provided; in addition, the candidate recommended friends can be determined according to the personal data information of the user, so that the coverage rate of recommended friends is increased, and the accuracy of user friend recommendation can be obviously increased.
Description
Technical field
The invention belongs to internet arena, particularly relate to a kind of friend recommendation method and apparatus.
Background technology
Along with developing rapidly of social networks, the customer group in social networks is also increasing.For the needs of user's Communication, social networking system usually can be recommended may interested good friend.
Existing friend recommendation method, general step comprises: determine the interested candidate crowd of user's possibility, sort to the carrying out of described candidate crowd, preferential recommendation sorts forward candidate crowd to user.Wherein, the method that candidate crowd sorts is comprised: sort based on common good friend's number and sort based on user property.
Existing friend recommendation method, can intelligence for user, recommend may interested good friend, but, because the influencing factor of the consideration when carrying out friend recommendation is comparatively single, make the accuracy of the good friend of recommendation not high.
Summary of the invention
The object of the embodiment of the present invention is a kind of method providing friend recommendation, comparatively single with the influencing factor solving prior art consideration when carrying out friend recommendation, makes the problem that the accuracy of the good friend recommended is not high.
The embodiment of the present invention is achieved in that a kind of friend recommendation method, and described method comprises:
Obtain the data message of user, the data message of described user comprises personal attribute information and the social networks topology information of user;
According to the data message of described user, determine the characteristic vector of the commending friends of described user candidate and the commending friends of described candidate, described characteristic vector comprises the quantitative information of the personal attribute of user and the quantitative information of social networks topological structure;
According to the model that training in advance obtains, calculate the score value that described characteristic vector is corresponding, obtain the score value sequence of user good friend;
Score value according to the user good friend calculated sorts, to user's commending friends.
Another object of the embodiment of the present invention is to provide a kind of friend recommendation device, and described device comprises:
Acquiring unit, for obtaining the data message of user, the data message of described user comprises personal attribute information and the social networks topology information of user;
Characteristic vector determining unit, for the data message according to described user, determine the characteristic vector of the commending friends of described user candidate and the commending friends of described candidate, described characteristic vector comprises the quantitative information of the personal attribute of user and the quantitative information of social networks topological structure;
Score value computing unit, for the model obtained according to training in advance, calculates the score value that described characteristic vector is corresponding, obtains the score value sequence of user good friend;
Friend recommendation unit, for the score value sequence according to the user good friend calculated, to user's commending friends.
In embodiments of the present invention, by considering the personal attribute information of user and social networks topology information, the characteristic vector of the commending friends of candidate and the commending friends of candidate is determined by the data message of user, the neural network model obtained by training in advance loads described characteristic vector, calculate the mark of the Candidate Recommendation good friend of user, can friend recommendation be carried out according to score.Owing to passing through personal attribute information and the topology network architecture information of synthetic user in the embodiment of the present invention, the reference data of recommendation is more comprehensive.In addition, determined the commending friends of candidate by the personal data information of user, make the coverage rate of commending friends more comprehensively, the accuracy rate of user's friend recommendation can be improved significantly.
Accompanying drawing explanation
Fig. 1 is the realization flow figure of the friend recommendation method that first embodiment of the invention provides;
Fig. 2 is the realization flow figure of the friend recommendation method that second embodiment of the invention provides;
Fig. 3 is the structured flowchart of the friend recommendation device that third embodiment of the invention provides;
Fig. 3 a for the invention process provide by the schematic diagram of friend recommendation application of installation in QQ Fellow searching;
The structured flowchart of the friend recommendation equipment that Fig. 4 provides for fourth embodiment of the invention.
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.
Friend recommendation method and apparatus described in the embodiment of the present invention, may be used for the application program with good friend's data such as immediate communication tool or social networks.
Method described in the embodiment of the present invention, mainly being can by conjunction with the personal attribute information of user and the social networks topology information of user, determine the scope of the good friend of user, two degree of good friends (good friend in the good friend contact person of user is passed through with existing, be called the once good friend of user, in the once good friend that user is all, the good friend of all once good friends, be called two degree of good friends) really the mode of the commending friends of candidate compare, improve the coverage of the commending friends of user candidate, and the topology information of user is in conjunction with userspersonal information, can the more efficiently accuracy improving user good friend data.
Embodiment one:
Fig. 1 shows the realization flow of a kind of friend recommendation method that first embodiment of the invention provides, and details are as follows:
In step S101, obtain the data message of user, the data message of described user comprises personal attribute information and the social networks topology information of user.
The data message of described acquisition user, can by obtaining the data such as off-line chat record of the personal information of user, private page or user, the age, sex, home address, work address, interest, occupation etc. of user is comprised at personal information, can also by obtaining user good friend to the evaluation information etc. of user; The interactive information of good friend can be comprised in private page, the interactive information of described good friend comprises the content module of the message content of other user in the page, access times to the reply content of described message, other user, access time and access, can by adding up the interactive information obtaining corresponding good friend to the page of personal homepage; Described off-line chat record comprises the frequency etc. that predetermined keyword occurs, the search by carrying out keyword to the content of chat record obtains the data message of described off-line chat record.
The data message of described user comprises personal attribute information and the social networks topology information of user, namely described in going up, by the personal information of user, the personal attribute information of user can be obtained, by the off-line chat record of private page or user, the social networks topology information of user can be obtained.
Wherein, as one comparatively preferred embodiment in, the social networks topology information of described user comprises following information: user with candidate commending friends belonging to whether the type of circle, the number of circle, user newly add good friend, whether the commending friends of candidate be that user newly adds the group friend of group, user with common good friend's number of the commending friends of candidate; The personal attribute information of described user comprises following information: whether user identical with the commending friends sex of candidate, user with the age gap of the commending friends of candidate, user with the commending friends of candidate address whether in same province or city, the gap of mean age of other gap of evenness of the commending friends of candidate and the good friend of user, the commending friends of candidate and the good friend of user.Certainly, as in the mode that may implement, the social networks topology information of described user and the personal attribute information of user, can comprise above-mentioned wherein one or more.
In step s 102, according to the data message of described user, determine the characteristic vector of the commending friends of described user candidate and the commending friends of described candidate, described characteristic vector comprises the quantitative information of the personal attribute of user and the quantitative information of social networks topological structure.
Wherein, determine the commending friends of user candidate according to the data message of user, two degree of good friends of user can be comprised, other user profile that the social networks topology information of user comprises.In the interactive information of the private page that the topology information of such as social networks comprises, the content module information etc. of the access times of the page, message and return information, concern, the similarity etc. of the key word information of the user that such as off-line chat record comprises.
Determine the characteristic vector of the commending friends of described candidate according to the data message of described user, specifically can comprise the standardizing step of the quantification to data message.
Described quantitative information, refers to the process of the situation of change of the data message of user by digital representation.
Quantization step can the first possible feature that comprises of statistical information, then all possible and that feature is corresponding value is defined, such as the age of user, can to possible feature, by stages defines, be 3 as differed value corresponding to 0-5 year with the age of user, the value differing 5-15 year with the age of user is 2, and the value differing more than 15 years old with age of user is 1.Other quantification is similar, illustrates no longer one by one.
In addition, as the execution mode that the embodiment of the present invention is optimized further, the weight coefficient of adding portion vector characteristics can also be comprised.The part vector characteristics adding weight coefficient comprises the change of address information or adds new good friend or the frequency of group.Thus can the needs of new interpolation good friend that bring of the change of monitor user ' environmental information timely and effectively.
In step s 103, according to the neural network model that training in advance obtains, calculate the score value that described characteristic vector is corresponding, obtain the score value sequence of user good friend.
According to the neural network model that training in advance obtains, load corresponding characteristic vector, replace the parameter of the correspondence in neural network model, thus obtain the value of the neural network model after alternative parameter.
Wherein, described neural net (being also called BP (English full name is Back Propagation) network) model is a kind of Multi-layered Feedforward Networks by Back Propagation Algorithm training.BP network can learn and store a large amount of input-output mode map relations, is constantly adjusted the weights and threshold of network by backpropagation, makes the error of network minimum.
In step S104, the score value according to the user good friend calculated sorts, to user's commending friends.
Loaded the characteristic vector calculating value of the commending friends of the candidate of user respectively by neural network model after, according to the size of value, sort from big to small, and according to sort recommendations good friend.
Wherein, carry out friend recommendation according to sequence, can recommend according to the threshold value of the value of neural network model, the commending friends of candidate when being greater than the threshold value of specifying to value is recommended, or also can according to the number of recommending, selected and sorted specifies the good friend of number to recommend above.
The embodiment of the present invention is by considering the personal attribute information of user and social networks topology information, the characteristic vector of the commending friends of candidate and the commending friends of candidate is determined by the data message of user, the neural network model obtained by training in advance loads described characteristic vector, calculate the mark of the Candidate Recommendation good friend of user, can friend recommendation be carried out according to score.Owing to passing through personal attribute information and the topology network architecture information of synthetic user in the embodiment of the present invention, the reference data of recommending is more comprehensive, and the commending friends of candidate is determined by the personal data information of user, make the coverage rate of commending friends more comprehensively, the accuracy rate of user's friend recommendation can be improved significantly.
Embodiment two:
Fig. 2 shows the realization flow of the friend recommendation method that second embodiment of the invention provides, and details are as follows:
In step s 201, obtain the data message of user, the data message of described user comprises personal attribute information and the social networks topology information of user.
Wherein, optionally, the social networks topology information of described user comprise following one or more: user with candidate commending friends belonging to whether the type of circle, the number of circle, user newly add good friend, whether the commending friends of candidate be that user newly adds the group friend of group, user with common good friend's number of the commending friends of candidate.
Optionally, the personal attribute information of described user comprise following one or more: whether user identical with the commending friends sex of candidate, user with the age gap of the commending friends of candidate, user with the commending friends of candidate address whether in same province or city, the gap of mean age of other gap of evenness of the commending friends of candidate and the good friend of user, the commending friends of candidate and the good friend of user.
In step S202, according to the data message of described user, determine the characteristic vector of the commending friends of described user candidate and the commending friends of described candidate, described characteristic vector comprises the quantitative information of the personal attribute of user and the quantitative information of social networks topological structure.
Wherein, determine the commending friends of described user candidate, can determine according to following one or several:
1, two degree of good friends in user's friend relation chain;
Contact person in the address list of 2, user communication terminal;
3, the interactive information in the individual subscriber page.
Wherein, the communication number of the contact person in the address list of described user communication terminal, can bind with immediate communication tool or social tool, when inquiring the communication number of contact person, by the corresponding relation in binding data storehouse, the number of immediate communication tool corresponding to contact person or the number of social tool can be inquired.
Interactive information in described private page, comprises multidate information, access private page number of times or the concern private page etc. that forward or issue in comment private page.
In order to better improve the accuracy of recommendation, at the described data message according to described user, before determining the characteristic vector step of the commending friends of described user candidate and the commending friends of described candidate, described method also comprises:
According to the improper keywords database preset, the data message filtering described user comprises the good friend of the Candidate Recommendation of the keyword in described improper keywords database.
For keyword in the improper keywords database that user data information comprises, can comprise as " intermediary ", " agency ", " advertisement " etc., wherein, described keyword search the keyword that can comprise at the name of contact person in address list, interactive information, interactive information such as in the page comprises above-mentioned keyword, the malicious user of that a part of sending advertisement can be made to filter.Thus improve efficiency and the accuracy of friend recommendation calculating.
In step S203, obtain the training sample data of m user, described training sample data comprise the characteristic vector of the commending friends of each user candidate, the target sequence y of friend recommendation
(i), 0<i<m+1, m are positive integer.
Concrete, the training sample data of m user, can by choosing in the data of historic user, the characteristic vector of the commending friends of candidate wherein, and the obtain manner according to the characteristic vector in enforcement one obtains.The target sequence of described friend recommendation, namely in history recommending data, the data of successful referral.Namely described successful referral represents that the good friend of recommendation obtains the accreditation of user, is initiatively initiated to add request by user.The target sequence of described friend recommendation, also comprises the order of successful referral, and the sequencing adding good friend according to user corresponds to the target sequence of friend recommendation.
The training sample data of described m, comprise m user, each user is to the commending friends that should have candidate, each Candidate Recommendation good friend and m user comprise its corresponding personal attribute information and social networks topology information, about the definition of personal attribute information and social networks topology information, it is identical with embodiment one.
In step S204, by the characteristic vector of the commending friends of candidate as the input amendment of neural network model, the target sequence of friend recommendation is as the output sample of neural network model, and training obtains neural network model.
Concrete, by the characteristic vector of the commending friends of candidate as the input amendment of neural network model, the target sequence of friend recommendation is as the output sample of neural network model, training obtains the mode of neural network model, can comprise multiple, such as existing mean square deviation minimum value method, the invention process provides a kind of algorithm being more better than LMS least mean square method, comprises the following steps:
Obtain the collating sequence z of the commending friends of the candidate of the user of stochastic generation
(i);
Calculate
the mark of corresponding characteristic vector, wherein, y
(i)for the target sequence of friend recommendation, goal ordering sequences y
(i)with randomly ordered sequence z
(i)cross entropy be expressed as:
Obtain weight coefficient corresponding to described characteristic vector according to the mark of described characteristic vector, the weight coefficient corresponding according to described characteristic vector generates neural network model.
Wherein, described in
Be a probability-distribution function, according to the positive negative sample of following structure: collect the recommending data in sample data, positive sample be designated to the commending friends of the candidate that those have request to add, do not ask the commending friends of the candidate adding good friend to be designated negative sample.So under this probability function, positive sample sequence is forward, and its probable value is higher, thus can better portray sequence effect.
Wherein, described calculating
in the numerical steps of corresponding characteristic vector, obtained by gradient descent method
the numerical value of corresponding characteristic vector.
Described gradient descent method, usually also referred to as steepest descent method, is referred to that the decrease speed of utility function in negative gradient direction is the fastest, is searched the method be worth most of function by this direction.
In step S205, according to the neural network model that training in advance obtains, calculate the score value that described characteristic vector is corresponding, obtain the score value sequence of user good friend.
By number of training according to this and the adjustment of cross entropy revised, the mark that each characteristic vector is corresponding is obtained.According to the specification quantized in advance, corresponding weight coefficient can be obtained.
Be exemplified below: in the quantizing process for the age of user, differing value corresponding to 0-5 year with the age of user is 3, the value differing 5-15 year with the age of user is 2, the value differing more than 15 years old with age of user is 1, calculating corresponding mark when to differ 0-5 year with the age of user is 0.3, is so 0.1 for differing weight coefficient corresponding to 0-5 year with the age of user.The rest may be inferred for other.
In step S206, the score value according to the user good friend calculated sorts, to user's commending friends.
In embodiments of the present invention, can be recommended when searching by immediate communication tool or social tool, or further recommend afterwards in interpolation, or can recommend in knowable people's page.
The embodiment of the present invention is compared with embodiment one, difference part is that the cross entropy by calculating target sequence and randomly ordered sequence in sample data is minimum, thus obtain weight coefficient corresponding to each characteristic vector, this computational methods can be taken into consideration the personal attribute information of user and social networks topology information simultaneously, and can be good at portraying sequence effect.In addition, by carrying out the screening of keyword to the commending friends of candidate, can prevent from recommending malicious user by mistake, improve the accuracy rate of friend recommendation.
Embodiment three:
Fig. 3 shows the structural representation of the friend recommendation device that third embodiment of the invention provides, and details are as follows:
The device of friend recommendation described in the embodiment of the present invention, comprising:
Acquiring unit 301, for obtaining the data message of user, the data message of described user comprises personal attribute information and the social networks topology information of user;
Characteristic vector determining unit 302, for the data message according to described user, determine the characteristic vector of the commending friends of described user candidate and the commending friends of described candidate, described characteristic vector comprises the quantitative information of the personal attribute of user and the quantitative information of social networks topological structure;
Score value computing unit 303, for the neural network model obtained according to training in advance, calculates the score value that described characteristic vector is corresponding, obtains the score value sequence of user good friend;
Friend recommendation unit 304, for the score value sequence according to the user good friend calculated, to user's commending friends.
Further, described device also can comprise:
Sample data acquiring unit, for obtaining the training sample data of m user, described training sample data comprise the characteristic vector of the commending friends of each user candidate, the target sequence y of friend recommendation
(i), 0<i<m+1, m are positive integer;
Training unit, for the characteristic vector of the commending friends by candidate as the input amendment of neural network model, the target sequence of friend recommendation is as the output sample of neural network model, and training obtains neural network model.
Concrete, described training unit comprises:
Collating sequence generates subelement, for obtaining the collating sequence z of the commending friends of the candidate of the user of stochastic generation
(i);
Computation subunit, for calculating
the mark of corresponding characteristic vector, wherein, goal ordering sequences y
(i)with randomly ordered sequence z
(i)cross entropy be expressed as:
Generate subelement, for obtaining weight coefficient corresponding to described characteristic vector according to the mark of described characteristic vector, the weight coefficient corresponding according to described characteristic vector generates neural network model.
Concrete, described computation subunit is specifically for obtaining by gradient descent method
the numerical value of corresponding characteristic vector.
For improving the coverage rate of the commending friends of candidate, in the execution mode optimized further, described device also comprises:
The commending friends determining unit of candidate, for determining the commending friends of user candidate according to following one or more: the contact person in two degree of good friends in user's friend relation chain, the address list of user communication terminal, the interactive information in the individual subscriber page.
Concrete, the social networks topology information of described user comprise following one or more: user with candidate commending friends belonging to whether the type of circle, the number of circle, user newly add good friend, whether the commending friends of candidate be that user newly adds the group friend of group, user with common good friend's number of the commending friends of candidate; The personal attribute information of described user comprise following one or more: whether user identical with the commending friends sex of candidate, user with the age gap of the commending friends of candidate, user with the commending friends of candidate address whether in same province or city, the gap of mean age of other gap of evenness of the commending friends of candidate and the good friend of user, the commending friends of candidate and the good friend of user.
For improving computational efficiency further and improving accuracy rate, described device also can comprise:
Filter element, for the improper keywords database that basis presets, the data message filtering described user comprises the good friend of the Candidate Recommendation of the keyword in described improper keywords database.
Fig. 3 a for the invention process provide by the schematic diagram of friend recommendation application of installation in QQ Fellow searching.
As shown in Figure 3 a, acquiring unit accesses the data of multiple data source, and comprise the off-line data of QQ good friend, the contact library of address list of QQ place mobile phone, the user having a talk about comment forwarding in QQ space, these data are the basis and information source of recommending.
Described characteristic vector determining unit receives the multiple data sources obtained from acquiring unit, in order to improve the accuracy rate of Fellow searching in data source, can also carry out preliminary screening to the commending friends of the candidate in data source.The commending friends of candidate comprises the two degree of good friends (i.e. the good friend of user good friend) in QQ friend relation chain, the contact person at the address list of user QQ place mobile phone, QQ space repeating or comments on the crowd of the original model of user.A part in the good friend of these candidates may be that user is interested.Contact person for example in user mobile phone is likely that they are interested, often forwards or comment on the model etc. that some people sends out in QQ space.In the commending friends of the candidate obtained, by beyond good friend's circle, the commending friends that topological structure contacts untight candidate filters, for mobile phone contact and comment content, can be filtered by keyword, as keyword lookup such as conventional " intermediaries ", " agency ", filter the commending friends of the uninterested candidate of those users.
By obtaining the commending friends of the candidate of user after screening, searching the feature of the commending friends of described candidate, and each feature unification is analyzed on same dimension, each feature quantized and analyzes.Described feature can comprise the size etc. as the number of the circle type of the commending friends of common good friend's number, user and candidate and circle, whether same sex, age gap distance.In addition, for the new plusing good friend of user or newly to add group be that user adds that of other good friends is important inspires condition, accordingly to this two category features weighting, it is allowed better to play a role.
Described score value computing unit, to screening and the feature after quantizing, calculates corresponding score value by order models.
The acquisition process of order models described in the embodiment of the present invention, is described in detail in embodiment two, obtains by statistical history recommending data the order models calculating score value.
The ranking score that described friend recommendation unit calculates according to score value computing unit, the sequence forming sequence exports, the commending friends of the candidate that preferential recommendation ranking score is high, the QQ that the outlet of the sequence of output comprises pc client searches recommendation, the QQ of pc client add good friend after recommendation, mobile phone QQ client asynchronization panel recommend, the pc client of QQ or can knowable people recommending of cell-phone customer terminal.
Friend recommendation device described in the embodiment of the present invention is corresponding with friend recommendation method described in embodiment one, embodiment two, does not repeat at this.
Embodiment four:
Fig. 4 shows that fourth embodiment of the invention provides
The structured flowchart of the equipment that Fig. 4 provides for fourth embodiment of the invention, equipment described in the present embodiment, comprising: the parts such as memory 420, processor 480 and power supply 490.It will be understood by those skilled in the art that the device structure shown in Fig. 4 does not form the restriction to equipment, the parts more more or less than diagram can be comprised, or combine some parts, or different parts are arranged.
Concrete introduction is carried out below in conjunction with Fig. 4 each component parts to equipment:
Memory 420 can be used for storing software program and module, and processor 480 is stored in software program and the module of memory 420 by running, thus the various function application of actuating equipment and data processing.Memory 420 mainly can comprise storage program district and store data field, and wherein, storage program district can storage operation system, application program (such as sound-playing function, image player function etc.) etc. needed at least one function; Store data field and can store the data (such as voice data, phone directory etc.) etc. created according to the use of equipment.In addition, memory 420 can comprise high-speed random access memory, can also comprise nonvolatile memory, such as at least one disk memory, flush memory device or other volatile solid-state parts.
Processor 480 is control centres of equipment, utilize the various piece of various interface and the whole equipment of connection, software program in memory 420 and/or module is stored in by running or performing, and call the data be stored in memory 420, the various function of actuating equipment and deal with data, thus integral monitoring is carried out to equipment.Optionally, processor 480 can comprise one or more processing unit; Preferably, processor 480 accessible site application processor and modem processor, wherein, application processor mainly processes operating system, user interface and application program etc., and modem processor mainly processes radio communication.Be understandable that, above-mentioned modem processor also can not be integrated in processor 480.
Equipment also comprises the power supply 490 (such as battery) of powering to all parts, preferably, power supply can be connected with processor 480 logic by power-supply management system, thus realizes the functions such as management charging, electric discharge and power managed by power-supply management system.
Although not shown, equipment can also comprise camera, bluetooth module, mixed-media network modules mixed-media, input module, display module etc., does not repeat them here.
In embodiments of the present invention, the processor 480 included by this equipment also has following functions: the method performing friend recommendation, specifically comprises:
Obtain the data message of user, the data message of described user comprises personal attribute information and the social networks topology information of user;
According to the data message of described user, determine the characteristic vector of the commending friends of described user candidate and the commending friends of described candidate, described characteristic vector comprises the quantitative information of the personal attribute of user and the quantitative information of social networks topological structure;
According to the neural network model that training in advance obtains, calculate the score value that described characteristic vector is corresponding, obtain the score value sequence of user good friend;
Score value according to the user good friend calculated sorts, to user's commending friends.
The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any amendments done within the spirit and principles in the present invention, equivalent replacement and improvement etc., all should be included within protection scope of the present invention.
Claims (14)
1. a friend recommendation method, is characterized in that, described method comprises:
Obtain the data message of user, the data message of described user comprises personal attribute information and the social networks topology information of user;
According to the data message of described user, determine the characteristic vector of the commending friends of described user candidate and the commending friends of described candidate, described characteristic vector comprises the quantitative information of the personal attribute of user and the quantitative information of social networks topological structure;
According to the neural network model that training in advance obtains, calculate the score value that described characteristic vector is corresponding, obtain the score value sequence of user good friend;
Score value according to the user good friend calculated sorts, to user's commending friends.
2. method according to claim 1, it is characterized in that, at the described neural network model obtained according to training in advance, calculate the score value that described characteristic vector is corresponding, before obtaining the score value ordered steps of user good friend, described method also comprises:
Obtain the training sample data of m user, described training sample data comprise the characteristic vector of the commending friends of each user candidate, the target sequence y of friend recommendation
(i), 0<i<m+1, m are positive integer;
By the characteristic vector of the commending friends of candidate as the input amendment of neural network model, the target sequence of friend recommendation is as the output sample of neural network model, and training obtains neural network model.
3. method according to claim 2, it is characterized in that, the described characteristic vector by the commending friends of candidate is as the input amendment of neural network model, and the target sequence of friend recommendation is as the output sample of neural network model, and training obtains neural network model step and comprises:
Obtain the collating sequence z (i) of the commending friends of the candidate of the user of stochastic generation;
Calculate
the mark of corresponding characteristic vector, wherein, goal ordering sequences y
(i)with randomly ordered sequence z
(i)cross entropy be expressed as:
Obtain weight coefficient corresponding to described characteristic vector according to the mark of described characteristic vector, the weight coefficient corresponding according to described characteristic vector generates neural network model.
4. method according to claim 3, is characterized in that, described calculating
in the numerical steps of corresponding characteristic vector, obtained by gradient descent method
the numerical value of corresponding characteristic vector.
5. method according to claim 1, is characterized in that, the described data message according to described user, determines that the commending friends step of described user candidate comprises:
The commending friends of user candidate is determined according to following one or more:
Contact person in two degree of good friends in user's friend relation chain, the address list of user communication terminal, the interactive information in the individual subscriber page.
6. method according to claim 1, it is characterized in that, the social networks topology information of described user comprise following one or more: user with candidate commending friends belonging to whether the type of circle, the number of circle, user newly add good friend, whether the commending friends of candidate be that user newly adds the group friend of group, user with common good friend's number of the commending friends of candidate; The personal attribute information of described user comprise following one or more: whether user identical with the commending friends sex of candidate, user with the age gap of the commending friends of candidate, user with the commending friends of candidate address whether in same province or city, the gap of mean age of other gap of evenness of the commending friends of candidate and the good friend of user, the commending friends of candidate and the good friend of user.
7. method according to claim 1, it is characterized in that, the described data message according to described user, before determining the characteristic vector step of the commending friends of described user candidate and the commending friends of described candidate, described method also comprises:
According to the improper keywords database preset, the data message filtering described user comprises the good friend of the Candidate Recommendation of the keyword in described improper keywords database.
8. a friend recommendation device, is characterized in that, described device comprises:
Acquiring unit, for obtaining the data message of user, the data message of described user comprises personal attribute information and the social networks topology information of user;
Characteristic vector determining unit, for the data message according to described user, determine the characteristic vector of the commending friends of described user candidate and the commending friends of described candidate, described characteristic vector comprises the quantitative information of the personal attribute of user and the quantitative information of social networks topological structure;
Score value computing unit, for the neural network model obtained according to training in advance, calculates the score value that described characteristic vector is corresponding, obtains the score value sequence of user good friend;
Friend recommendation unit, for the score value sequence according to the user good friend calculated, to user's commending friends.
9. device according to claim 8, it is characterized in that, described device also comprises:
Sample data acquiring unit, for obtaining the training sample data of m user, described training sample data comprise the characteristic vector of the commending friends of each user candidate, the target sequence y of friend recommendation
(i), 0<i<m+1, m are positive integer;
Training unit, for the characteristic vector of the commending friends by candidate as the input amendment of neural network model, the target sequence of friend recommendation is as the output sample of neural network model, and training obtains neural network model.
10. device according to claim 9, it is characterized in that, described training unit comprises:
Collating sequence generates subelement, for obtaining the collating sequence z of the commending friends of the candidate of the user of stochastic generation
(i);
Computation subunit, for calculating
the mark of corresponding characteristic vector, wherein, goal ordering sequences y
(i)with randomly ordered sequence z
(i)cross entropy be expressed as:
Generate subelement, for obtaining weight coefficient corresponding to described characteristic vector according to the mark of described characteristic vector, the weight coefficient corresponding according to described characteristic vector generates neural network model.
11. methods according to claim 10, it is characterized in that, described computation subunit is specifically for obtaining by gradient descent method
the numerical value of corresponding characteristic vector.
12. devices according to claim 8, it is characterized in that, described device also comprises:
The commending friends determining unit of candidate, for determining the commending friends of user candidate according to following one or more: the contact person in two degree of good friends in user's friend relation chain, the address list of user communication terminal, the interactive information in the individual subscriber page.
13. devices according to claim 8, it is characterized in that, the social networks topology information of described user comprise following one or more: user with candidate commending friends belonging to whether the type of circle, the number of circle, user newly add good friend, whether the commending friends of candidate be that user newly adds the group friend of group, user with common good friend's number of the commending friends of candidate; The personal attribute information of described user comprise following one or more: whether user identical with the commending friends sex of candidate, user with the age gap of the commending friends of candidate, user with the commending friends of candidate address whether in same province or city, the gap of mean age of other gap of evenness of the commending friends of candidate and the good friend of user, the commending friends of candidate and the good friend of user.
14. devices according to claim 8, it is characterized in that, described device also comprises:
Filter element, for the improper keywords database that basis presets, the data message filtering described user comprises the good friend of the Candidate Recommendation of the keyword in described improper keywords database.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410180918.9A CN104601438B (en) | 2014-04-28 | 2014-04-28 | A kind of friend recommendation method and apparatus |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410180918.9A CN104601438B (en) | 2014-04-28 | 2014-04-28 | A kind of friend recommendation method and apparatus |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104601438A true CN104601438A (en) | 2015-05-06 |
| CN104601438B CN104601438B (en) | 2017-12-01 |
Family
ID=53126959
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410180918.9A Active CN104601438B (en) | 2014-04-28 | 2014-04-28 | A kind of friend recommendation method and apparatus |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104601438B (en) |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105245435A (en) * | 2015-09-28 | 2016-01-13 | 广州市百果园网络科技有限公司 | Friend recommending method and system |
| CN105574131A (en) * | 2015-12-14 | 2016-05-11 | 西安交通大学 | Social network friend making recommendation method and system based on dynamic community identification |
| CN105843830A (en) * | 2015-11-30 | 2016-08-10 | 维沃移动通信有限公司 | Friend recommendation method and system |
| CN106021325A (en) * | 2016-05-06 | 2016-10-12 | 腾讯科技（深圳）有限公司 | A friend recommendation method and device |
| CN106446198A (en) * | 2016-09-29 | 2017-02-22 | 北京百度网讯科技有限公司 | Recommending method and device of news based on artificial intelligence |
| CN106997347A (en) * | 2016-01-22 | 2017-08-01 | 华为技术有限公司 | Information recommendation method and server |
| CN107123057A (en) * | 2017-03-22 | 2017-09-01 | 阿里巴巴集团控股有限公司 | User recommends method and device |
| CN107368579A (en) * | 2017-07-21 | 2017-11-21 | 佛山潮伊汇服装有限公司 | Social user recommends method |
| CN107809370A (en) * | 2016-09-09 | 2018-03-16 | 腾讯科技（深圳）有限公司 | User recommends method and device |
| CN108683734A (en) * | 2018-05-15 | 2018-10-19 | 广州虎牙信息科技有限公司 | Category method for pushing, device and storage device, computer equipment |
| CN108769125A (en) * | 2018-04-28 | 2018-11-06 | 广州优视网络科技有限公司 | Using recommendation method, apparatus, storage medium and computer equipment |
| CN108875993A (en) * | 2017-05-16 | 2018-11-23 | 清华大学 | Invite behavior prediction method and device |
| CN109062914A (en) * | 2017-06-12 | 2018-12-21 | 东软集团股份有限公司 | User's recommended method and device, storage medium and server |
| CN109062900A (en) * | 2018-08-01 | 2018-12-21 | 科大讯飞股份有限公司 | A kind of circle of friends generation method and device |
| CN109101614A (en) * | 2018-08-06 | 2018-12-28 | 百度在线网络技术（北京）有限公司 | Friend recommendation method and apparatus |
| CN109299388A (en) * | 2018-11-16 | 2019-02-01 | 杭州安芯科技有限公司 | A kind of system and method for searching high quality social activity user |
| CN109446171A (en) * | 2017-08-30 | 2019-03-08 | 腾讯科技（深圳）有限公司 | A kind of data processing method and device |
| CN109785858A (en) * | 2018-12-14 | 2019-05-21 | 平安普惠企业管理有限公司 | A kind of contact person's adding method, device, readable storage medium storing program for executing and terminal device |
| CN109872242A (en) * | 2019-01-30 | 2019-06-11 | 北京字节跳动网络技术有限公司 | Information-pushing method and device |
| CN110992196A (en) * | 2019-12-13 | 2020-04-10 | 北京达佳互联信息技术有限公司 | Social relationship establishing method and system, terminal device and server |
| CN111353093A (en) * | 2018-12-24 | 2020-06-30 | 北京嘀嘀无限科技发展有限公司 | Question recommendation method and device, server and readable storage medium |
| CN111368219A (en) * | 2020-02-27 | 2020-07-03 | 广州腾讯科技有限公司 | Information recommendation method and device, computer equipment and storage medium |
| CN111861635A (en) * | 2020-06-17 | 2020-10-30 | 北京邮电大学 | Friend recommendation method, device and equipment for commodity sharing |
| CN112132656A (en) * | 2020-09-21 | 2020-12-25 | 深圳思为科技有限公司 | Guest judging method and related equipment |
| CN112532512A (en) * | 2020-12-16 | 2021-03-19 | 深圳市镜玩科技有限公司 | Session message processing method, related device, equipment and medium |
| CN112633901A (en) * | 2020-12-18 | 2021-04-09 | 深圳市思为软件技术有限公司 | Guest judging method and related equipment |
| CN114423051A (en) * | 2021-12-22 | 2022-04-29 | 中国电信股份有限公司 | Information recommendation method and device, electronic equipment and computer readable medium |
| CN114677234A (en) * | 2022-04-26 | 2022-06-28 | 河南大学 | Graph convolution neural network social contact recommendation method and system integrating multi-channel attention mechanism |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020416A (en) * | 2011-09-26 | 2013-04-03 | 北京千橡网景科技发展有限公司 | Method and equipment for friend recommendation in multiuser online games |
| CN103150595A (en) * | 2011-12-06 | 2013-06-12 | 腾讯科技（深圳）有限公司 | Automatic pair selection method and device in data processing system |
| CN103455515A (en) * | 2012-06-01 | 2013-12-18 | 腾讯科技（深圳）有限公司 | User recommendation method and system in SNS (social networking services) community |
-
2014
- 2014-04-28 CN CN201410180918.9A patent/CN104601438B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020416A (en) * | 2011-09-26 | 2013-04-03 | 北京千橡网景科技发展有限公司 | Method and equipment for friend recommendation in multiuser online games |
| CN103150595A (en) * | 2011-12-06 | 2013-06-12 | 腾讯科技（深圳）有限公司 | Automatic pair selection method and device in data processing system |
| CN103455515A (en) * | 2012-06-01 | 2013-12-18 | 腾讯科技（深圳）有限公司 | User recommendation method and system in SNS (social networking services) community |
Cited By (46)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105245435A (en) * | 2015-09-28 | 2016-01-13 | 广州市百果园网络科技有限公司 | Friend recommending method and system |
| CN105245435B (en) * | 2015-09-28 | 2018-11-06 | 广州市百果园网络科技有限公司 | A kind of friend recommendation method and system |
| CN105843830A (en) * | 2015-11-30 | 2016-08-10 | 维沃移动通信有限公司 | Friend recommendation method and system |
| CN105574131A (en) * | 2015-12-14 | 2016-05-11 | 西安交通大学 | Social network friend making recommendation method and system based on dynamic community identification |
| CN105574131B (en) * | 2015-12-14 | 2017-04-26 | 西安交通大学 | Social network friend making recommendation method and system based on dynamic community identification |
| CN106997347A (en) * | 2016-01-22 | 2017-08-01 | 华为技术有限公司 | Information recommendation method and server |
| CN106021325A (en) * | 2016-05-06 | 2016-10-12 | 腾讯科技（深圳）有限公司 | A friend recommendation method and device |
| CN107809370A (en) * | 2016-09-09 | 2018-03-16 | 腾讯科技（深圳）有限公司 | User recommends method and device |
| CN106446198A (en) * | 2016-09-29 | 2017-02-22 | 北京百度网讯科技有限公司 | Recommending method and device of news based on artificial intelligence |
| CN107123057A (en) * | 2017-03-22 | 2017-09-01 | 阿里巴巴集团控股有限公司 | User recommends method and device |
| 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 |
| CN109062914B (en) * | 2017-06-12 | 2020-10-23 | 东软集团股份有限公司 | User recommendation method and device, storage medium and server |
| CN109062914A (en) * | 2017-06-12 | 2018-12-21 | 东软集团股份有限公司 | User's recommended method and device, storage medium and server |
| CN107368579A (en) * | 2017-07-21 | 2017-11-21 | 佛山潮伊汇服装有限公司 | Social user recommends method |
| CN109446171A (en) * | 2017-08-30 | 2019-03-08 | 腾讯科技（深圳）有限公司 | A kind of data processing method and device |
| CN109446171B (en) * | 2017-08-30 | 2022-03-15 | 腾讯科技（深圳）有限公司 | Data processing method and device |
| CN108769125A (en) * | 2018-04-28 | 2018-11-06 | 广州优视网络科技有限公司 | Using recommendation method, apparatus, storage medium and computer equipment |
| CN108769125B (en) * | 2018-04-28 | 2021-08-17 | 阿里巴巴（中国）有限公司 | Application recommendation method and device, storage medium and computer equipment |
| CN108683734B (en) * | 2018-05-15 | 2021-04-09 | 广州虎牙信息科技有限公司 | Method and device for pushing classes, storage equipment and computer equipment |
| CN108683734A (en) * | 2018-05-15 | 2018-10-19 | 广州虎牙信息科技有限公司 | Category method for pushing, device and storage device, computer equipment |
| CN109062900B (en) * | 2018-08-01 | 2023-01-17 | 科大讯飞股份有限公司 | Friend circle generation method and device |
| CN109062900A (en) * | 2018-08-01 | 2018-12-21 | 科大讯飞股份有限公司 | A kind of circle of friends generation method and device |
| CN109101614A (en) * | 2018-08-06 | 2018-12-28 | 百度在线网络技术（北京）有限公司 | Friend recommendation method and apparatus |
| CN109299388A (en) * | 2018-11-16 | 2019-02-01 | 杭州安芯科技有限公司 | A kind of system and method for searching high quality social activity user |
| CN109785858A (en) * | 2018-12-14 | 2019-05-21 | 平安普惠企业管理有限公司 | A kind of contact person's adding method, device, readable storage medium storing program for executing and terminal device |
| CN109785858B (en) * | 2018-12-14 | 2024-02-23 | 深圳市兴海物联科技有限公司 | Contact person adding method and device, readable storage medium and terminal equipment |
| CN111353093A (en) * | 2018-12-24 | 2020-06-30 | 北京嘀嘀无限科技发展有限公司 | Question recommendation method and device, server and readable storage medium |
| CN111353093B (en) * | 2018-12-24 | 2023-05-23 | 北京嘀嘀无限科技发展有限公司 | Problem recommendation method, device, server and readable storage medium |
| WO2020156389A1 (en) * | 2019-01-30 | 2020-08-06 | 北京字节跳动网络技术有限公司 | Information pushing method and device |
| CN109872242A (en) * | 2019-01-30 | 2019-06-11 | 北京字节跳动网络技术有限公司 | Information-pushing method and device |
| CN110992196B (en) * | 2019-12-13 | 2023-12-12 | 北京达佳互联信息技术有限公司 | Social relation establishment method, system, terminal equipment and server |
| CN110992196A (en) * | 2019-12-13 | 2020-04-10 | 北京达佳互联信息技术有限公司 | Social relationship establishing method and system, terminal device and server |
| CN111368219A (en) * | 2020-02-27 | 2020-07-03 | 广州腾讯科技有限公司 | Information recommendation method and device, computer equipment and storage medium |
| CN111368219B (en) * | 2020-02-27 | 2024-04-26 | 广州腾讯科技有限公司 | Information recommendation method, device, computer equipment and storage medium |
| CN111861635B (en) * | 2020-06-17 | 2022-10-25 | 北京邮电大学 | Friend recommendation method, device and equipment for commodity sharing |
| CN111861635A (en) * | 2020-06-17 | 2020-10-30 | 北京邮电大学 | Friend recommendation method, device and equipment for commodity sharing |
| CN112132656A (en) * | 2020-09-21 | 2020-12-25 | 深圳思为科技有限公司 | Guest judging method and related equipment |
| CN112132656B (en) * | 2020-09-21 | 2024-05-14 | 深圳思为科技有限公司 | Passenger judging method and related equipment |
| CN112532512B (en) * | 2020-12-16 | 2023-03-17 | 深圳市镜玩科技有限公司 | Session message processing method, related device, equipment and medium |
| CN112532512A (en) * | 2020-12-16 | 2021-03-19 | 深圳市镜玩科技有限公司 | Session message processing method, related device, equipment and medium |
| CN112633901A (en) * | 2020-12-18 | 2021-04-09 | 深圳市思为软件技术有限公司 | Guest judging method and related equipment |
| CN114423051B (en) * | 2021-12-22 | 2024-04-30 | 中国电信股份有限公司 | Information recommendation method, device, electronic equipment and computer readable medium |
| CN114423051A (en) * | 2021-12-22 | 2022-04-29 | 中国电信股份有限公司 | Information recommendation method and device, electronic equipment and computer readable medium |
| CN114677234A (en) * | 2022-04-26 | 2022-06-28 | 河南大学 | Graph convolution neural network social contact recommendation method and system integrating multi-channel attention mechanism |
| CN114677234B (en) * | 2022-04-26 | 2024-04-30 | 河南大学 | Graph convolution neural network social recommendation method and system integrating multichannel attention mechanisms |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104601438B (en) | 2017-12-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104601438A (en) | Friend recommendation method and device | |
| CN110472107B (en) | Multi-mode knowledge graph construction method, device, server and storage medium | |
| TW201901539A (en) | Assessment method of risk, device, computer device and storage medium | |
| CN102279889B (en) | A kind of question pushing method and system based on geography information | |
| CN104486129B (en) | The method and system of application service quality are ensured under distributed environment | |
| CN105224959A (en) | The training method of order models and device | |
| CN113326377A (en) | Name disambiguation method and system based on enterprise incidence relation | |
| CN107306355A (en) | A kind of content recommendation method and server | |
| CN112819263B (en) | Method and device for scheduling customer service | |
| CN110555172A (en) | user relationship mining method and device, electronic equipment and storage medium | |
| CN113254472A (en) | Parameter configuration method, device, equipment and readable storage medium | |
| CN103942298B (en) | Recommendation method and system based on linear regression | |
| CN111970335A (en) | Information recommendation method and device and storage medium | |
| CN110287424B (en) | Collaborative filtering recommendation method based on single-source SimRank | |
| WO2015165230A1 (en) | Social contact message monitoring method and device | |
| CN111008873A (en) | User determination method and device, electronic equipment and storage medium | |
| CN115700548A (en) | Method, apparatus and computer program product for user behavior prediction | |
| CN111931035B (en) | Service recommendation method, device and equipment | |
| CN110705889A (en) | Enterprise screening method, device, equipment and storage medium | |
| CN113034231B (en) | Multi-supply chain commodity intelligent recommendation system and method based on SaaS cloud service | |
| CN111562990B (en) | Lightweight serverless computing method based on message | |
| CN114818672A (en) | Text duplicate removal method and device, electronic equipment and readable storage medium | |
| Pourheidari et al. | Recommender system based on extracted data from different social media. a study of twitter and linkedin | |
| Duan et al. | Towards building and evaluating a personalized location-based recommender system | |
| CN103853711A (en) | Text information processing method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |