WO2015027828A1 - 实现用户信息聚类的方法和装置 - Google Patents

实现用户信息聚类的方法和装置 Download PDF

Info

Publication number
WO2015027828A1
WO2015027828A1 PCT/CN2014/084484 CN2014084484W WO2015027828A1 WO 2015027828 A1 WO2015027828 A1 WO 2015027828A1 CN 2014084484 W CN2014084484 W CN 2014084484W WO 2015027828 A1 WO2015027828 A1 WO 2015027828A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
range
location
setting information
information
Prior art date
Application number
PCT/CN2014/084484
Other languages
English (en)
French (fr)
Inventor
马腾
吴瑕
Original Assignee
腾讯科技(深圳)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 腾讯科技(深圳)有限公司 filed Critical 腾讯科技(深圳)有限公司
Publication of WO2015027828A1 publication Critical patent/WO2015027828A1/zh

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/01Social networking
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/10Office automation; Time management

Definitions

  • the present invention relates to information processing techniques, and more particularly to a method and apparatus for implementing clustering of user information. Background technique
  • Accessing social networks through social applications has gradually become the mainstream of users' network access. For example, users initiate or participate in various party activities in social networks through social applications.
  • the initiation and participation of any activity requires manual setting of the involved users, for example, the user who initiated the activity manually specifies the active user, or the user applies to participate in the activity after browsing to the activity initiated in the social network. Activities to aggregate several users among a large number of users.
  • this method is inefficient due to the need for manual confirmation by the user, and it is difficult to achieve aggregation of user characteristics among a large number of users. Summary of the invention
  • a method for implementing clustering of user information including the following steps:
  • a clustering result is generated based on the set of users and a desired range.
  • an apparatus for implementing user information clustering including: a clustering module, configured to quantize user information to obtain corresponding user features, and cluster the corresponding user features to obtain a user set;
  • a range statistic module configured to obtain range setting information of the user set, and perform statistics on the range setting information to obtain a desired range corresponding to the user set;
  • a result generating module configured to generate a clustering result according to the set of users and a desired range.
  • the foregoing method and device for realizing user information clustering are performed according to user information to obtain corresponding user features, and clustering corresponding user features to obtain a user set, so as to implement user aggregation among a large number of users for each user information, and obtain a user. Collection, the users in the user collection have the same or similar user characteristics.
  • the users in the user set are also statistically calculated according to the range setting information to obtain the expected range corresponding to the user set, and then the clustering result is generated according to the user set and the expected range, thereby realizing The prediction of user behavior is such that the generated clustering results are reasonable and accurate for the users in the user collection.
  • FIG. 1 is a flowchart of a method for implementing user information clustering in an embodiment
  • FIG. 2 is a schematic diagram of a curve of raw data in one embodiment
  • FIG. 3 is a schematic diagram of a spectral space formed by the original data in FIG. 2;
  • FIG. 4 is a flow chart of a method for constructing a spectral space according to user information, performing Laplacian feature mapping to obtain a user's vertices in the spectral space, and performing user clustering on the vertices of the spectral space to obtain a user set. ;
  • FIG. 5 is a flow chart of a method for quantizing user information to construct a similarity matrix in FIG. 4;
  • FIG. 6 is a flowchart of a method for obtaining range setting information and performing statistics on range setting information to obtain a desired range corresponding to a user set in an embodiment
  • FIG. 7 is a flowchart of a method for obtaining range setting information and performing statistics on range setting information to obtain a desired range corresponding to a user set in another embodiment
  • FIG. 8 is a flow chart of a method for acquiring location range setting information in FIG. 7; 9 is a flow chart of a method for dividing a location range in a location range setting information into a plurality of location sub-ranges in FIG. 8;
  • FIG. 10 is a schematic structural diagram of an apparatus for implementing user information clustering in an embodiment
  • FIG. 11 is a schematic structural diagram of a clustering module in FIG. 10;
  • Figure 12 is a schematic structural view of the quantizing unit of Figure 11;
  • FIG. 13 is a schematic structural diagram of a range statistics module in an embodiment
  • FIG. 15 is a schematic structural diagram of a second information acquiring unit in FIG. 14;
  • Figure 16 is a block diagram showing the structure of the second dividing unit of Figure 14. detailed description
  • a method for implementing clustering of user information includes the following steps:
  • step S10 the user information is quantized to obtain corresponding user features, and the corresponding user features are clustered to obtain a user set.
  • the user information includes basic information such as the user's age, gender, hobbies, and the like.
  • the user information is quantized and clustered to obtain a plurality of similar users, and a plurality of users are formed to form a user set.
  • step S10 is: constructing a spectral space according to user information, and performing a Laplacian feature mapping to obtain a vertice of the user in the spectral space, and the user is clustered by the user at the vertices of the spectral space to obtain the user. set.
  • the spectral clustering theory is based on the spectral theory in graph theory, and its essence is to transform the clustering problem into the optimal cutting problem of the graph.
  • the spectral clustering algorithm can divide the sample space of arbitrary shape and converge to the global optimal solution.
  • the original data with high similarity in the spectral space is concentrated, while the data with low similarity is scattered.
  • the original data is two spiral curves.
  • the appropriate cluster space is constructed according to the eigenvalues and eigenvectors of the spectrum, that is, the spectral space, as shown in Fig. 3, in the spectral space. in, The sampling points on the different curves are divided into two piles to perform accurate clustering on this basis.
  • Step S30 Acquire range setting information, and perform statistics on the range setting information to obtain a desired range of the user set.
  • the range setting information includes time range setting information and location range setting information, wherein the time range setting information is a union formed by a time range of each user in the user set; the location range setting information is a user The union formed by the range of locations for each user in the collection.
  • the range setting information can be used to know the range conditions delineated by the user set, and in this range condition, the sub-range accepted by the user, that is, the expected range, is obtained for the users in the user set.
  • Step S50 generating a clustering result according to the user set and the expected range.
  • a clustering result including a user set and a desired range is generated, and the user included in the user set and the corresponding expected range are known according to the clustering result.
  • the object of the participation activity that is, the user in the user collection, and the time range and location range of the activity implementation can be learned through the clustering result, thereby avoiding users composed of multiple users.
  • the complex process of group discussion of time and place of activity and the lack of efficiency of communication caused by the unification of opinions have improved the speed of information processing in social networks.
  • the user can also dynamically recommend the clustering result of the activity that can be initiated for the user who accesses the social network.
  • the user can view the clustering result to know the current set of users that can initiate the activity and the expected range, and then according to the clustering result. Initiating an event can greatly improve the convenience of offline activities in social networks.
  • the above clustering results can be implemented through virtual social network tools and portals provided in social applications such as instant messaging tools, or corresponding portal implementations can be added to the electronic map, and can also be set as independent applications, and the generated clustering results will be Push to social apps, e-maps, or other stand-alone apps for users to view.
  • the foregoing constructs a spectral space according to user information, and performs Laplacian feature mapping to obtain a vertice of the user in the spectral space, and the user is clustered by the user at the vertices of the spectral space to obtain a user.
  • the steps of the collection include the following steps: In step S110, the user information is quantized to construct a similarity matrix.
  • the user information may be obtained by the registration information of the user, or may be input by the user.
  • the user information is quantized by dimension to obtain a quantized value corresponding to each dimension in each user information. For example, in the user information, age and gender correspond to one dimension.
  • Step S130 extracting a Laplacian matrix from the similarity matrix, and performing feature decomposition on the Laplacian matrix to construct a spectral space W.
  • the Laplacian matrix of the spectral space W is L.
  • the Lapp component block is diagonally formed, namely:
  • the spectral space W is opened by m eigenvectors, and the m eigenvectors are corresponding to 0 eigenvalues.
  • Step S150 mapping the quantized user information to the spectral space to obtain the vertices of the user in the spectral space.
  • is the number of clusters.
  • the position of 1 indicates the partition to which the vertex belongs, meaning that these points are mapped to the same point in the spectral space.
  • points belonging to the same block ⁇ are mapped to different points in the spectral space.
  • the points projected into the spectral space will satisfy: Similar points will be closer, dissimilar points will be farther away, and similar points will occur when the disturbance is not large. It is mapped to a more concentrated position in the spectral space, so the perturbation factor will also be considered, and the user corresponding to the points mapped to the spectral space is determined by K-means clustering to obtain the user ⁇ A ⁇ o
  • Step S170 forming a user set by the user corresponding to the vertices that are concentrated on each other in the spectral space.
  • the users corresponding to the vertices gathered together in the spectral space are obtained, that is, the points at which the plurality of users are mapped to the spectral space are grouped together, and the users are formed into a user set.
  • the foregoing step S110 includes:
  • Step S111 construct a vector corresponding to each user according to the user information.
  • Step S113 normalize the elements in the vector, and perform weighted calculation on the normalized elements to obtain a quantized value corresponding to the user.
  • the elements in the vector are normalized according to the dimension corresponding to the element, and the weighted calculation is performed to obtain the quantized value corresponding to the user.
  • Step S115 calculating a distance between the quantized values corresponding to the user to obtain a similarity between the two users, and forming an adjacency matrix by the similarity between the two users.
  • the weight is ⁇ [(U] , which describes the similarity between ⁇ and Vj .
  • Q Q .
  • the similarity matrix W is the adjacency matrix of the undirected weighted graph G, the corresponding adjacency matrix will be constructed to truly reflect the similarity relationship between the vertex ⁇ and Vj .
  • the distance between the quantized values corresponding to the user that is, the distance between I.i, is calculated to obtain the similarity between the two users, and the adjacency matrix is formed by the similarity.
  • Step S117 obtaining a similarity matrix by the adjacency matrix.
  • the similar matrix is obtained from the calculated adjacency matrix.
  • the range setting information is time range setting information; and the step S30 includes:
  • step S310a time range setting information is acquired.
  • time range setting information in which the union of time ranges corresponding to all users in the user set is recorded is obtained.
  • the time range setting information may be obtained by taking a union of acceptable time ranges input for each user in the set of users.
  • step S330a the time range in the time range setting information is divided into a plurality of time sub-intervals.
  • the earliest time and the latest time in the acceptable time range of all users are extracted from the time range setting information, and the earliest time and the latest time constitute the time range T in the time range setting information.
  • the time range ⁇ is divided into ⁇ time subintervals with time intervals ⁇ / ⁇ .
  • Step S350a The user in the user set is counted as the number of acceptable times of the user corresponding to each time subinterval, and the time subinterval whose user accepts the maximum number of times is used as the time expectation range corresponding to the user set.
  • a counter of length n is used to count each time subinterval in the time range T to obtain an acceptable number of times corresponding to each time subinterval.
  • the endpoint of the line segment will indicate the start and end times acceptable to the user.
  • the intersection of multiple collinear segments will be reduced to the intersection of two collinear segments, and it is judged whether the two collinear segments intersect. It is only necessary to determine whether the endpoint of one segment is between another segment. Therefore, in the process of calculating the time expectation range, since it is impossible to ensure that the two line segments must intersect, it is only necessary to calculate the optimal solution, and therefore, the calculation process is simply and quickly calculated by the above calculation process.
  • the range setting information is location range setting information.
  • mobile devices can be used to perform GPS (Global Positioning System) positioning to obtain the user's current location and behavioral trajectory. Therefore, in order to obtain a range of locations acceptable to the user, the current location of the user is obtained, and the current location of the user is taken as the center, and the radius corresponding to the user is obtained according to the traffic mode currently used by the user, and the circle is the unit time.
  • the above step S30 includes:
  • Step S310b acquiring location range setting information.
  • the location range setting information records the union of the circles corresponding to the users in the user set, and the union is the location range.
  • step S310b includes the following steps:
  • Step S311b Acquire positioning information corresponding to the user in the user set.
  • the positioning information obtained by the GPS positioning or input by the user is obtained, and then the current location of the user can be obtained according to the positioning information.
  • Step S313b Determine a location range corresponding to the user according to the location in the location information, and obtain a location range setting information by combining the location ranges corresponding to the user.
  • the traffic mode or the current speed adopted by the user is obtained, and the radius corresponding to the user is obtained according to the current traffic mode or current speed used by the user, and then the corresponding circle is obtained according to the position and radius in the positioning information.
  • the circle is the location range corresponding to the user.
  • the circle of all the users in the user collection is obtained, and the location range of all users in the location range setting information is obtained.
  • the location range in the location range setting information is divided into a plurality of location sub-ranges.
  • the division location is in a plurality of parts to obtain a plurality of location sub-ranges.
  • step S330b includes the following steps:
  • Step S331b according to the location range layout icon in the location range setting information.
  • the location range is presented in the form of a diagram to facilitate more intuitive division of the location range.
  • step S333b the illustration is divided into a plurality of grids, which are the location sub-ranges.
  • the illustration is meshed to obtain a grid of m' n , and each grid is a location sub-range.
  • step S350b the user in the user set counts the number of users corresponding to each sub-range of the location, and the sub-range of the location where the user accepts the maximum number of times is used as the location expectation range corresponding to the user set.
  • the counter sub-range is counted by using a counter C of length m' n, and the element d in the obtained array is the acceptable number of users of the i-th sub-range, and the location corresponding to the element with the largest value.
  • the sub-range is the expected range of the location corresponding to the user set.
  • an apparatus for implementing clustering of user information includes a clustering module 10, a range statistics module 30, and a result generation module 50.
  • the clustering module 10 is configured to quantize the user information to obtain corresponding user features, and cluster the features corresponding to the user to obtain the user set.
  • the user information includes basic information such as the user's age, gender, hobbies, and the like.
  • the user information is quantized and clustered to obtain a plurality of similar users, and a plurality of users are formed to form a user set.
  • the clustering module 10 is further configured to construct a spectral space according to user information, and perform a Laplacian feature mapping to obtain a vertice of the user in the spectral space, and the user is clustered by the user at the vertices of the spectral space. User collection.
  • the spectral clustering theory is based on the spectral theory in graph theory, and its essence is to transform the clustering problem into the optimal cutting problem of the graph.
  • the spectral clustering algorithm can divide the sample space of arbitrary shape and converge to the global optimal solution.
  • the original data with high similarity in the spectral space is concentrated, while the data with low similarity is scattered.
  • the range statistics module 30 is configured to acquire the range setting information of the user, and perform statistics on the range setting information. To get the desired range corresponding to the user set.
  • the range statistics module 30 in order to accurately pre-determine and rationalize the user, the range statistics module 30 also obtains additional range setting information to set a reasonable expected range for the user set, so that the users in the user set are similar. And the expected range is also consistent with the development of user behavior and user-related events in the user collection.
  • the range setting information includes time range setting information and location range setting information, wherein the time range setting information is a union formed by a time range of each user in the user set; the location range setting information The union formed by the range of locations for each user in the user collection.
  • the range statistic module 30 can know the range condition delineated by the user set through the scoping setting information, and then in this range condition, the sub-range accepted by the user, that is, the expected range, is obtained for the user in the user set.
  • the result generating module 50 is configured to generate a clustering result according to the user set and the expected range.
  • the result generating module 50 generates a clustering result including the user set and the expected range, and according to the clustering result, the user included in the user set and the corresponding expected range are obtained.
  • the object of the participation activity that is, the user in the user collection, and the time range and location range of the activity implementation can be learned through the clustering result, thereby avoiding users composed of multiple users.
  • the complex process of group discussion of time and place of activity and the lack of efficiency of communication caused by the unification of opinions have improved the speed of information processing in social networks.
  • the user can also dynamically recommend the clustering result of the activity that can be initiated for the user who accesses the social network.
  • the user can view the clustering result to know the current set of users that can initiate the activity and the expected range, and then according to the clustering result. Initiating an event can greatly improve the convenience of offline activities in social networks.
  • the above clustering results can be implemented through virtual social network tools and portals provided in social applications such as instant messaging tools, or corresponding portal implementations can be added to the electronic map, and can also be set as independent applications, and the generated clustering results will be Push to social apps, e-maps, or other stand-alone apps for users to view.
  • the clustering module 10 includes a quantization unit 110, a spectral space construction unit 130, a mapping unit 150, and a set forming unit 170.
  • the quantization unit 110 is configured to quantize the user information to construct a similarity matrix.
  • the user information may be obtained by the user's registration information, or may be a user input.
  • the quantization unit uo quantizes the dimension in the user information to obtain a quantized value corresponding to each dimension in each user information. For example, in the user information, age and gender correspond to one dimension.
  • the spectral space construction unit 130 is configured to extract a Laplacian matrix from the similarity matrix, and perform feature decomposition on the Laplacian matrix to construct a spectral space.
  • the spectral space construction unit 130 calculates a Laplacian matrix from the similarity matrix. According to the connected part of each point in the Laplacian matrix, the Laplacian matrix L is written into the diagonal form of the block, gp :
  • the spectral space W is opened by m eigenvectors, and these m eigenvectors are corresponding to 0 eigenvalues. .
  • the mapping unit 150 is configured to map the quantized user information to the spectral space to obtain the vertices of the user in the spectral space.
  • the set forming unit determines the user corresponding to the same point mapped in the spectral space by K-means clustering to obtain the user set.
  • the set forming unit 170 is configured to form a user set corresponding to the vertices that are concentrated on each other in the spectral space.
  • the set forming unit 170 acquires users corresponding to the vertices that are concentrated together in the spectral space, that is, the points at which the plurality of users are mapped to the spectral space are grouped together, and the users constitute the user ⁇ A A ⁇ o
  • the quantization unit 110 includes a vector construction unit 111, an operation unit 113, a similarity calculation unit 115, and a similarity matrix acquisition unit 117.
  • the vector construction unit 111 is configured to construct a vector corresponding to each user according to the user information.
  • the vector building unit U1 will use the n-dimensional vector I to describe P".
  • Each element in the vector I will represent a kind of user information in the user. Dimensions.
  • the operation unit 113 is configured to perform normalization processing on the elements in the vector, and perform weighted calculation on the normalized elements to obtain a quantized value corresponding to the user.
  • the operation unit U3 normalizes the elements in the vector according to the dimension corresponding to the element, and performs weighting calculation to obtain the quantized value corresponding to the user.
  • the similarity calculation unit 115 is configured to calculate a distance between the quantized values corresponding to the user to obtain a similarity between the two users, and form an adjacency matrix by the similarity between the two users.
  • the similarity calculating unit 115 constructs a corresponding adjacency matrix to truly reflect the similarity relationship between the vertex ⁇ and Vj .
  • the similarity calculation unit 115 calculates the distance between the quantized values corresponding to the user, that is, the distance between L and L, to obtain the similarity ⁇ between the two users, and further forms the adjacent matrix by the similarity ⁇ .
  • the similarity matrix acquisition unit 117 is configured to obtain a similarity matrix by the adjacency matrix.
  • the similar matrix obtaining unit U7 obtains a similar matrix from the calculated adjacency matrix.
  • the range setting information is time range setting information
  • the range statistics module 30 includes a first information acquiring unit 310a, a first dividing unit 330a, and a first sub-interval counting unit 350a. .
  • the first information acquiring unit 310a is configured to acquire time range setting information.
  • the first information acquiring unit 310a acquires time range setting information that records the union of time ranges corresponding to all users in the user set.
  • the time range setting information may be an input through input to each user in the user set.
  • the time range received is taken from the union.
  • the first dividing unit 330a is configured to divide the time range in the time range setting information into a plurality of time sub-intervals.
  • the first dividing unit 330a extracts the earliest time and the latest time among all the acceptable time ranges of the user from the time range setting information, and the earliest time and the latest time constitute the time range setting information.
  • the time range T divides the time range ⁇ into n time subintervals with time intervals ⁇ / ⁇ .
  • the first sub-interval statistic unit 350a is configured to count the number of acceptable times of the user corresponding to each time sub-interval for the user in the user set, and use the time sub-interval with the largest number of acceptable times of the user as the time expectation range corresponding to the user set.
  • the first sub-interval statistic unit 350a uses a counter of length n to count each time sub-interval in the time range T to obtain an acceptable user for each time sub-interval.
  • the above calculation process according to the expected range of time, it is converted into the intersection of multiple collinear segments.
  • the endpoint of the line segment will indicate the start and end times acceptable to the user.
  • the intersection of multiple collinear segments will be reduced to the intersection of two collinear segments, and judging whether the two collinear segments intersect is only necessary to determine whether the endpoint of one segment is between another segment. Therefore, in the process of calculating the time desired range, since it is impossible to ensure that the two line segments must intersect, it is only necessary to calculate the optimal solution, and therefore, the calculation process is simply and quickly calculated by the above calculation process.
  • the range setting information is location range setting information.
  • mobile devices such as smart terminals and in-vehicle mobile terminals
  • mobile devices can be used for GPS (Global)
  • the range statistic module 30 includes a second information acquiring unit 310b, a second dividing unit 330b, and a second sub-interval statistic unit 350b.
  • the second information acquiring unit 310b is configured to acquire location range setting information.
  • the location range setting information records the circle corresponding to the user in the user set.
  • Set, the union is the location range.
  • the second information acquiring unit 310b includes a positioning information acquiring unit 311b and a position determining unit 313b.
  • the location information obtaining unit 311b is configured to acquire positioning information corresponding to the user in the user set.
  • the location information acquiring unit 311b obtains the location information obtained by the GPS positioning or input by the user, and then the location where the user is currently located can be obtained according to the location information.
  • the location determining unit 313b is configured to determine a location range corresponding to the user according to the location in the location information, and combine the location ranges corresponding to the user to obtain location range setting information.
  • the location determining unit 313b obtains the traffic mode or the current speed adopted by the user, so as to obtain the radius corresponding to the user according to the current traffic mode or current speed used by the user, and then according to the location and radius in the positioning information.
  • a corresponding circle is obtained, and the circle is the location range corresponding to the user.
  • the circle of all users in the user collection is obtained and the location range of all users in the location range setting information is obtained.
  • the second dividing unit 330b is configured to divide the location range in the location range setting information into a plurality of location sub-ranges.
  • the second dividing unit 330b divides the location range into a plurality of parts to obtain a plurality of location sub-ranges.
  • the second dividing unit 330b includes a routing unit 331b and a mesh dividing unit 333b.
  • the layout unit 331b is configured to arrange the map according to the location range in the location range setting information.
  • the layout unit 331b allows the location range to be present in the form of a diagram to facilitate more intuitive division of the location range.
  • a mesh dividing unit 333b is used to divide the illustration into a plurality of grids, which is a location sub-range.
  • the mesh dividing unit 333b performs meshing on the icons to obtain a mesh of m'n, and each mesh is a spot sub-range.
  • the second sub-interval statistic unit 350b is configured to count the number of acceptable times of the user corresponding to each sub-range of the user in the user set, and use the sub-range of the location where the user accepts the maximum number of times as the location desired range corresponding to the user collection.
  • the second sub-interval statistic unit 350b uses the counter C of length m'n to the location.
  • the sub-range is counted, and the elements in the obtained array are the acceptable number of users of the i-th sub-range, and the sub-range corresponding to the element with the largest value is the desired range of the location corresponding to the user set.
  • the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Primary Health Care (AREA)
  • Data Mining & Analysis (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明提供了一种实现用户信息聚类的方法和装置。所述方法包括:量化用户信息得到相应的用户特征,聚类所述用户特征得到用户集合;获取用户集合的范围设定信息,对所述范围设定信息进行统计以得到用户集合的期望范围;根据所述用户集合和期望范围生成聚类结果。所述装置包括:聚类模块、范围统计模块和结果生成模块。

Description

实现用户信息聚类的方法和装置
相关申请交叉引用
本申请要求 2013年 8月 26日提交中国专利局、申请号为 201310376923. 2、 发明名称为 "实现用户信息聚类的方法和装置" 的中国专利申请的优先权, 其 全部内容通过引用结合在本申请中。 技术领域
本发明涉及信息处理技术, 特别是涉及一种实现用户信息聚类的方法和装 置。 背景技术
随着虚拟社交网络工具以及即时通信工具 (诸如语音聊天、 图片分享) 等 各种社交应用的爆发式的增长和发展,, 交友的便利性得到了大大地提高。
通过社交应用访问社交网络已经逐渐成为用户进行网络访问的主流, 例如, 用户通过社交应用在社交网络中发起或参加各种聚会活动。 然而, 在社交网络 中, 任一活动的发起和参与都需要人工设置所涉及的用户, 例如, 发起活动的 用户人工指定活动的用户, 或者用户在浏览到社交网络中发起的活动之后申请 参与该活动, 以在海量的用户中将若干个用户聚合在一起, 然而, 这一方式由 于需要用户进行人工确认, 因此效率低下, 也难于实现海量用户中针对用户特 性的聚合。 发明内容
基于此, 根据本申请的实施例, 提供了一种实现用户信息聚类的方法和装 置。
根据本申请的一方面, 提供了一种实现用户信息聚类的方法, 包括如下步 骤:
量化用户信息得到对应的特征, 聚类所述相应的用户特征得到用户集合; 获取用户集合的范围设定信息, 对所述范围设定信息进行统计以得到用户 集合所对应的期望范围;
根据所述用户集合和期望范围生成聚类结果。
根据本申请的另一方面, 提供了一种实现用户信息聚类的装置, 包括: 聚类模块, 用于量化用户信息得到相应的用户特征, 聚类所述相应的用户 特征得到用户集合;
范围统计模块, 用于获取用户集合的范围设定信息, 对所述范围设定信息 进行统计以得到用户集合所对应的期望范围;
结果生成模块, 用于根据所述用户集合和期望范围生成聚类结果。
上述实现用户信息聚类的方法和装置, 根据用户信息进行量化得到相应的 用户特征, 聚类相应的用户特征得到用户集合, 以针对每一用户的用户信息在 海量用户中实现用户聚合, 得到用户集合, 该用户集合中的用户具备了相同或 相近似的用户特性。 此外, 为保证聚类的准确性, 还将根据范围设定信息对用 户集合中的用户进行统计以得到该用户集合所对应的期望范围, 进而根据用户 集合和期望范围生成聚类结果, 实现了对用户行为的预测, 以使得生成的聚类 结果对于用户集合中的用户而言是合理且准确的。 附图说明
图 1为一个实施例中实现用户信息聚类的方法流程图;
图 2为一个实施例中原始数据的曲线示意图;
图 3为图 2中原始数据形成的谱空间示意图;
图 4为图一个实施例中根据用户信息构造谱空间, 并进行拉普拉斯特征映 射得到用户在谱空间的顶点, 通过用户在谱空间的顶点对用户进行聚类得到用 户集合的方法流程图;
图 5为图 4中量化用户信息以构造相似矩阵的方法流程图;
图 6为一个实施例中获取范围设定信息, 对范围设定信息进行统计以得到 用户集合所对应的期望范围的方法流程图;
图 7为另一个实施例中获取范围设定信息, 对范围设定信息进行统计以得 到用户集合所对应的期望范围的方法流程图;
图 8为图 7中获取地点范围设定信息的方法流程图; 图 9为图 8中将地点范围设定信息中的地点范围划分为若干个地点子范围 的方法流程图;
图 10为一个实施例中实现用户信息聚类的装置的结构示意图;
图 11为图 10中聚类模块的结构示意图;
图 12为图 11中量化单元的结构示意图;
图 13为一个实施例中范围统计模块的结构示意图;
图 14为一个实施例中范围统计模块的结构示意图;
图 15为图 14中第二信息获取单元的结构示意图;
图 16为图 14中第二划分单元的结构示意图。 具体实施方式
为了使本发明的目的、 技术方案及优点更加清楚明白, 以下结合附图及实 施例, 对本发明进行进一步详细说明。 应当理解, 此处所描述的具体实施例仅 仅用以解释本发明, 并不用于限定本发明。
如图 1 所示, 在一个实施例中, 一种实现用户信息聚类的方法, 包括如下 步骤:
步骤 S10,量化用户信息得到相应的用户特征, 聚类相应的用户特征得到用 户集合。
本实施例中, 用户信息包括了用户的年龄、 性别、 兴趣爱好等基本信息。 对用户信息进行量化和聚类处理以得到相近似的多个用户, 进而由得到的多个 用户形成用户集合。
在一个实施例中, 上述步骤 S10 的过程为: 根据用户信息构造谱空间, 并 进行拉普拉斯特征映射得到用户在谱空间的顶点, 通过用户在谱空间的顶点对 用户进行聚类得到用户集合。
本实施例中, 谱聚类理论是建立在图论中谱图理论基础上的, 其本质是将 聚类问题转化为图的最优切割问题。 谱聚类算法能够对任意形状的样本空间进 行划分, 且收敛于全局最优解, 相应的, 在谱空间中相似性高的原始数据分布 比较集中, 而相似性低的数据分布则比较分散。
如图 2 所示, 原始数据为两根螺旋状的曲线, 谱聚类理论中根据谱图的特 征值和特征向量构造合适的聚类空间, 即谱空间, 如图 3 所示, 在谱空间中, 不同曲线上的采样点被分成两堆, 以在此基础上进行准确聚类。
步骤 S30,获取范围设定信息, 对范围设定信息进行统计以得到用户集合的 期望范围。
本实施例中, 为了对用户进行准确预设和合理性评价, 还将获取额外的范 围设定信息为用户集合设定合理的期望范围, 以使得用户集合中的用户是相近 似的, 并且期望范围也是与用户集合中的用户行为以及用户相关事件的发展相 符的。
进一步的, 范围设定信息包括时间范围设定信息和地点范围设定信息, 其 中, 时间范围设定信息为用户集合中每一用户的时间范围所形成的并集; 地点 范围设定信息为用户集合中每一用户的位置范围所形成的并集。
通过范围设定信息可获知用户集合所划定的范围条件, 进而在这一范围条 件中统计得到对用户集合中的用户而言, 最多用户接受的子范围, 即期望范围。
步骤 S50,根据用户集合和期望范围生成聚类结果。
本实施例中, 生成包含了用户集合和期望范围的聚类结果, 根据该聚类结 果可获知用户集合中包含的用户以及相应的期望范围。
例如, 对于社交网络中发起活动的用户而言, 可通过聚类结果获知参与活 动的对象, 即用户集合中的用户, 以及活动实施的时间范围和地点范围, 避免 了多个用户所构成的用户群体进行活动时间和地点讨论的复杂过程以及各方意 见不统一而造成的沟通缺乏效率的情况, 提高了社交网络中信息处理的速度。
此外, 也可为访问社交网络的用户动态的推荐可发起活动的聚类结果, 用 户通过查看这一聚类结果即可获知当前可发起活动的用户集合以及期望范围, 进而依据这一聚类结果发起活动即可, 大大提高了社交网络中线下活动的便利 性。
上述聚类结果可通过虚拟社交网络工具以及即时通信工具等社交应用中提 供的入口实现, 也可以在电子地图中增设相应的入口实现, 还可以设置为独立 的应用, 所生成的聚类结果将推送至社交应用、 电子地图或者其它的独立应用 中, 以供用户查看。
如图 4所示, 在一个实施例中, 上述根据用户信息构造谱空间, 并进行拉 普拉斯特征映射得到用户在谱空间的顶点, 通过用户在谱空间的顶点对用户进 行聚类得到用户集合的步骤包括如下步骤: 步骤 S110,量化用户信息以构造相似矩阵。
本实施例中, 用户信息可以是由用户的注册信息得到的, 也可以是用户输 入的。 对用户信息中按照维度进行量化以得到每一用户信息中每一维度所对应 的量化数值。 例如, 用户信息中, 年龄和性别都分别对应一个维度。
步骤 S 130,由相似矩阵提取拉普拉斯矩阵, 对拉普拉斯矩阵进行特征分解 以构造谱空间 W。
本实施例中, 设谱空间 W的拉普拉斯矩阵为 L。根据拉普拉斯矩阵中各项点 所属的连通部分, 将拉普 成分块对角形式, 即:
Figure imgf000006_0001
设拉普拉斯矩阵 L中 0特征值的个数为 m,则谱空间 W由 m个特征向量张开, 设这些 m个特征向量是由 0特征值对应的。
步骤 S150,将量化的用户信息映射至谱空间得到用户在谱空间的顶点。
本实施例中, 设 '为谱空间所对应的矩阵中第 i行对应的向量, 则在拉普拉 斯矩阵中所有属于分块 的顶点 v '都有相同的形式, 即 ( ι' ομ,其中,
Κ为聚类的个数, 1的位置表明了该顶点所属的分块, 意味着这些点都被映射到 谱空间中的同一点。
此外, 由于扰动的存在, 属于同一个分块 ^的点会被映射到谱空间中的不 同点。 根据极化定理 (Polarization Theorem)可知, 投影到谱空间中的点将满 足: 相似的点会靠得更近, 不相似的点会离的更远, 并且在扰动不大时, 相似 的点会被映射到谱空间中比较集中的位置, 因此还将考虑扰动的因素, 通过 K-means聚类确定被映射到谱空间中的点集中在一起所对应的用户,以得到用户 隹朱 A Π o
步骤 S170,将谱空间中距离上相互集中的顶点所对应的用户形成用户集合。 本实施例中, 获取谱空间中集中在一起的顶点所对应的用户, 即多个用户 被映射到了谱空间中的点集中在一起, 将这些用户构成用户集合。
如图 5所示, 在一个实施例中, 上述步骤 S110包括:
步骤 S111,根据用户信息构建每一用户所对应的向量。
本实施例中, 设所有用户的集合为 Ρ = {ΡΡ^, 将使用 η维向量 I描述 , 向量 I中的每一元素将表示所在用户中用户信息的一种维度。
步骤 S113 , 对向量中的元素进行归一化处理, 并将归一化处理后的元素进 行加权计算得到用户对应的量化值。
本实施例中, 根据元素所对应的维度对向量中的元素进行归一化处理, 并 进行加权计算后得到用户所对应的量化值。
步骤 S115,计算用户对应的量化值之间的距离得到两个用户之间的相似度, 并通过两个用户之间的相似度形成邻接矩阵。
本实施例中, 对于所有用户的集合 P, 可采用无向加权图 G= (V,E)表 , 其 中, V' 表示点 , ^和^被边集合£中的一条边相连, 该边的权值为^ [(U], 描述了 ^和 Vj的相似度, 越大说明 ^和 ^越相似。 特别地 = Q
由于相似度矩阵 W为无向加权图 G的邻接矩阵, 因此, 将构建相应的邻接 矩阵, 以真实地反映顶点 ^和 Vj之间的相似关系。
具体的, 将计算用户对应的量化值之间的距离, 即 I.i之间的距离, 以 得到两个用户之间的相似度 ,进而由相似度 形成邻接矩阵。
步骤 S117,通过邻接矩阵得到相似矩阵。
本实施例中, 由计算得到的邻接矩阵相应得到相似矩阵。
如图 6所示, 在一个实施例中, 上述范围设定信息为时间范围设定信息; 上述步骤 S30包括:
步骤 S310a,获取时间范围设定信息。
本实施例中, 为得到用户集合中用户之间重叠频率最高的时间段, 将获取 记录了用户集合中所有用户对应的时间范围的并集的时间范围设定信息。
例如, 该时间范围设定信息可以是通过对用户集合中每一用户输入的可接 受的时间范围取并集得到的。
步骤 S330a,将时间范围设定信息中的时间范围划分为若干个时间子区间。 本实施例中, 从时间范围设定信息中提取得到所有用户可接受时间范围中 的最早时间和最晚时间, 该最早时间和最晚时间便构成了时间范围设定信息中 的时间范围 T。 将时间范围 Τ划分成 η个时间间隔为 Τ/η的时间子区间。
步骤 S350a,对用户集合中的用户统计每一时间子区间所对应的用户可接受 次数, 并将用户可接受次数最大的时间子区间作为用户集合对应的时间期望范 围。 本实施例中, 使用长度为 n的计数器, 对时间范围 T中每一时间子区间进 行计数, 以得到每个时间子区间所对应的用户可接受次数。
上述根据时间所进行的期望范围计算过程中, 将其转化为多条共线线段的 求交问题。 其中, 线段端点将表示用户可接受的起始时刻和终止时刻。 多条共 线线段的求交将被归结为两条共线线段的求交, 而判断两条共线线段是否相交 只有需要判断其中一条线段的端点是否在另一条线段之间即可。 因此, 在计算 得到时间期望范围的过程中, 由于不能够保证两条线段一定相交, 只需要计算 最优解即可, 因此, 将通过上述计算过程进行简易而快速地计算。
在一个实施例中, 上述范围设定信息为地点范围设定信息。 随着智能终端 以及车载移动终端等多种移动设备的普及, 可利用移动设备进行 GPS ( Global Positioning System,全球定位系统) 定位以得到用户当前所在位置以及行为轨 迹。 因此, 为得到用户可接受的地点范围, 将获取用户当前所在位置, 并以用 户当前所在位置为圆心, 依据用户当前采用的交通方式设置半径得到该用户所 对应的圆, 该圆即为单位时间内用户所能到达的范围。 所采用的交通方式的速 度越快, 则半径越大, 此时, 求解用户集合中用户最可接受的地点范围的问题 则被转化为多个圆的求交问题, 因此, 如图 7所示, 上述步骤 S30包括:
步骤 S310b,获取地点范围设定信息。
本实施例中, 地点范围设置信息中记录了用户集合中用户所对应的圆的并 集, 该并集即为地点范围。
如图 8所示, 在一个实施例中, 上述步骤 S310b包括如下步骤:
步骤 S311b,获取用户集合中用户所对应的定位信息。
本实施例中, 获取由 GPS定位得到的或者用户输入的定位信息, 进而根据 定位信息即可获知用户当前所在的位置。
步骤 S313b, 根据定位信息中的位置确定用户对应的位置范围, 对用户对应 的位置范围取并集得到地点范围设定信息。
本实施例中, 获取用户所采用的交通方式或者当前速度, 以根据用户当前 所采用的交通方式或者当前速度得到该用户对应的半径, 进而根据定位信息中 的位置和半径即可得到相应的圆, 该圆即为用户对应的位置范围, 此时, 将用 户集合中所有用户的圆进行求并就得到了地点范围设定信息中所有用户的地点 范围。 步骤 S330b,将地点范围设定信息中的地点范围划分为若干个地点子范围。 本实施例中, 划分地点范围为多个部分, 以得到若干个地点子范围。
如图 9所示, 在一个实施例中, 上述步骤 S330b包括如下步骤:
步骤 S331b,根据地点范围设定信息中的地点范围布设图示。
本实施例中, 使地点范围以图示的形式存在, 以便于更为直观地对地点范 围进行划分。
步骤 S333b,将图示划分为若干个网格, 该网格即为地点子范围。
本实施例中, 对图示进行网格划分, 以得到 m' n的网格, 每一网格即为一 个地点子范围。
步骤 S350b,对用户集合中的用户统计每一地点子范围所对应的用户可接受 次数, 并将用户可接受次数最大的地点子范围作为用户集合对应的地点期望范 围。
本实施例中, 使用长度为 m' n的计数器 C对地点子范围进行计数, 所得到 的数组中元素 d即为第 i个地点子范围的用户可接受次数,数值最大的元素所对 应的地点子范围即为用户集合对应的地点期望范围。
如图 10所示, 在一个实施例中, 一种实现用户信息聚类的装置, 包括聚类 模块 10、 范围统计模块 30和结果生成模块 50。
聚类模块 10, 用于量化用户信息得到相应的用户特征, 聚类用户对应的特 征得到用户集合。
本实施例中, 用户信息包括了用户的年龄、 性别、 兴趣爱好等基本信息。 对用户信息进行量化和聚类处理以得到相近似的多个用户, 进而由得到的 多个用户形成用户集合。
在一个实施例中, 上述聚类模块 10还用于根据用户信息构造谱空间, 并进 行拉普拉斯特征映射得到用户在谱空间的顶点, 通过用户在谱空间的顶点对用 户进行聚类得到用户集合。
本实施例中, 谱聚类理论是建立在图论中谱图理论基础上的, 其本质是将 聚类问题转化为图的最优切割问题。 谱聚类算法能够对任意形状的样本空间进 行划分, 且收敛于全局最优解, 相应的, 在谱空间中相似性高的原始数据分布 比较集中, 而相似性低的数据分布则比较分散。
范围统计模块 30,用于获取用户的范围设定信息,对范围设定信息进行统计 以得到用户集合所对应的期望范围。
本实施例中, 为了对用户进行准确预设和合理性评价, 范围统计模块 30还 将获取额外的范围设定信息为用户集合设定合理的期望范围, 以使得用户集合 中的用户是相近似的, 并且期望范围也是与用户集合中的用户行为以及用户相 关事件的发展相符的。
进一步的, 范围设定信息将包括了时间范围设定信息和地点范围设定信息, 其中, 时间范围设定信息为用户集合中每一用户的时间范围所形成的并集; 地 点范围设定信息为用户集合中每一用户的位置范围所形成的并集。
范围统计模块 30通过范围设定信息可获知用户集合所划定的范围条件, 进 而在这一范围条件中统计得到对用户集合中的用户而言, 最多用户接受的子范 围, 即期望范围。
结果生成模块 50, 用于根据用户集合和期望范围生成聚类结果。
本实施例中,结果生成模块 50生成包含了用户集合和期望范围的聚类结果, 根据该聚类结果可获知用户集合中包含的用户以及相应的期望范围。
例如, 对于社交网络中发起活动的用户而言, 可通过聚类结果获知参与活 动的对象, 即用户集合中的用户, 以及活动实施的时间范围和地点范围, 避免 了多个用户所构成的用户群体进行活动时间和地点讨论的复杂过程以及各方意 见不统一而造成的沟通缺乏效率的情况, 提高了社交网络中信息处理的速度。
此外, 也可为访问社交网络的用户动态的推荐可发起活动的聚类结果, 用 户通过查看这一聚类结果即可获知当前可发起活动的用户集合以及期望范围, 进而依据这一聚类结果发起活动即可, 大大提高了社交网络中线下活动的便利 性。
上述聚类结果可通过虚拟社交网络工具以及即时通信工具等社交应用中提 供的入口实现, 也可以在电子地图中增设相应的入口实现, 还可以设置为独立 的应用, 所生成的聚类结果将推送至社交应用、 电子地图或者其它的独立应用 中, 以供用户查看。
如图 11所示, 在一个实施例中, 上述聚类模块 10包括量化单元 110、 谱空 间构造单元 130、 映射单元 150和集合形成单元 170。
量化单元 110,用于量化用户信息以构造相似矩阵。
本实施例中, 用户信息可以是由用户的注册信息得到的, 也可以是用户输 入的。 量化单元 uo对用户信息中按照维度进行量化以得到每一用户信息中每 一维度所对应的量化数值。 例如, 用户信息中, 年龄和性别都分别对应一个维 度。
谱空间构造单元 130,用于由相似矩阵提取拉普拉斯矩阵, 对拉普拉斯矩阵 进行特征分解以构造谱空间。
本实施例中, 谱空间构造单元 130 由相似矩阵相应计算得到拉普拉斯矩阵 (Laplacian矩阵)。根据拉普拉斯矩阵中各项点所属的连通部分, 将拉普拉斯矩 阵 L写成分块对角形式, gp :
Figure imgf000011_0001
设拉普拉斯矩阵中 0特征值的个数为 m,则谱空间 W由 m个特征向量张开, 这 些 m个特征向量是由 0特征值对应的。。
映射单元 150,用于将量化的用户信息映射至谱空间得到用户在谱空间的顶 点。
本实施例中, 设 '为谱空间所对应的矩阵中第 i行对应的向量, 则在拉普拉 斯矩阵中所有属于分块 的顶点 v '都有相同的形式, 即 ( ^ ι'ο^'ομ,其中, κ为聚类的个数, 1的位置表明了该顶点所属的分块, 意味着这些点都被映射到 谱空间中的同一点。
此外, 由于扰动的存在, 属于同一个分块 ^的点会被映射到谱空间中的不 同点。 根据极化定理 (Polarization Theorem)可知, 投影到谱空间中的点将满 足: 相似的点会靠得更近, 不相似的点会离的更远, 并且在扰动不大时, 相似 的点会被映射到谱空间中比较集中的位置, 因此还将考虑扰动的因素, 集合形 成单元通过 K-means 聚类确定被映射到谱空间中的同一点所对应的用户, 以得 到用户集合。
集合形成单元 170,用于将谱空间中距离上相互集中的顶点所对应的用户形 成用户集合。
本实施例中, 集合形成单元 170获取谱空间中集中在一起的顶点所对应的 用户, 即多个用户被映射到了谱空间中的点集中在一起, 将这些用户构成用户 隹朱 A Π o 如图 12所示, 在一个实施例中, 上述量化单元 110包括向量构建单元 111、 运算单元 113、 相似性计算单元 115和相似矩阵获取单元 117。
向量构建单元 111,用于根据用户信息构建每一用户所对应的向量。
本实施例中, 设所有用户的集合为 ^ ^1'…^),向量构建单元 U 1将使用 n 维向量 I描述 P"向量 I中的每一元素将表示所在用户中用户信息的一种维度。
运算单元 113,用于对向量中的元素进行归一化处理, 并将归一化处理后的 元素进行加权计算得到用户对应的量化值。
本实施例中, 运算单元 U3根据元素所对应的维度对向量中的元素进行归 一化处理, 并进行加权计算后得到用户所对应的量化值。
相似性计算单元 115,用于计算用户对应的量化值之间的距离得到两个用户 之间的相似度, 并通过两个用户之间的相似度形成邻接矩阵。
本实施例中, 对于所有用户的集合 P, 可采用无向加权图 G= (V,E)表 , 其 中, V' 表示点 , ^和^被边集合£中的一条边相连, 该边的权值为^ [Q'1], 描述了 ^和 Vj的相似度, 越大说明 ^和 ^越相似。 特别地 = Q
由于相似度矩阵 W为无向加权图 G的邻接矩阵, 因此, 相似性计算单元 115 将构建相应的邻接矩阵, 以真实地反映顶点 ^和 Vj之间的相似关系。
具体的, 相似性计算单元 115将计算用户对应的量化值之间的距离, 即 L 和 L之间的距离,以得到两个用户之间的相似度^,进而由相似度^形成邻接矩 阵。
相似矩阵获取单元 117,用于通过邻接矩阵得到相似矩阵。
本实施例中, 相似矩阵获取单元 U7 由计算得到的邻接矩阵相应得到相似 矩阵。
如图 13所示, 在一个实施例中, 上述范围设定信息为时间范围设定信息, 上述范围统计模块 30包括第一信息获取单元 310a、 第一划分单元 330a和第一 子区间统计单元 350a。
第一信息获取单元 310a, 用于获取时间范围设定信息。
本实施例中, 为得到用户集合中用户之间重叠频率最高的时间段, 第一信 息获取单元 310a将获取记录了用户集合中所有用户对应的时间范围的并集的时 间范围设定信息。
例如, 该时间范围设定信息可以是通过对用户集合中每一用户输入的可接 受的时间范围取并集得到的。
第一划分单元 330a, 用于将时间范围设定信息中的时间范围划分为若干个 时间子区间。
本实施例中, 第一划分单元 330a从时间范围设定信息中提取得到所有用户 可接受时间范围中的最早时间和最晚时间, 该最早时间和最晚时间便构成了时 间范围设定信息中的时间范围 T,将时间范围 Τ划分成 η个时间间隔为 Τ/η的时 间子区间。
第一子区间统计单元 350a,用于对用户集合中的用户统计每一时间子区间 所对应的用户可接受次数, 并将用户可接受次数最大的时间子区间作为用户集 合对应的时间期望范围。
本实施例中, 第一子区间统计单元 350a使用长度为 n的计数器, 对时间范 围 T 中每一时间子区间进行计数, 以得到每个时间子区间所对应的用户可接受
、 、'上述根据时间所进行的期望范围计算过程中, 将其转化为多条共线线段的 求交问题。 其中, 线段端点将表示用户可接受的起始时刻和终止时刻。 多条共 线线段的求交将被归结为两条共线线段的求交, 而判断两条共线线段是否相交 只需要判断其中一条线段的端点是否在另一条线段之间即可。 因此, 在计算得 到时间期望范围的过程中, 由于不能够保证两条线段一定相交, 只需要计算最 优解即可, 因此, 将通过上述计算过程进行简易而快速地计算。
在一个实施例中, 上述范围设定信息为地点范围设定信息。 随着智能终端 以及车载移动终端等多种移动设备的普及, 可利用移动设备进行 GPS ( Global
Positioning System,全球定位系统) 定位以得到用户当前所在位置以及行为轨 迹。 因此, 为得到用户可接受的地点范围, 将获取用户当前所在位置, 并以用 户当前所在位置为圆心, 依据用户当前采用的交通方式设置半径得到该用户所 对应的圆, 该圆即为单位时间内用户所能到达的范围。 所采用的交通方式的速 度越快, 则半径越大, 此时, 求解用户集合中用户最可接受的地点范围的问题 则被转化为多个圆的求交问题, 因此, 如图 14所示, 上述范围统计模块 30包 括第二信息获取单元 310b、 第二划分单元 330b和第二子区间统计单元 350b。
第二信息获取单元 310b,用于获取地点范围设定信息。
本实施例中, 地点范围设置信息中记录了用户集合中用户所对应的圆的并 集, 该并集即为地点范围。
如图 15所示, 在一个实施例中, 上述第二信息获取单元 310b包括定位信 息获取单元 311b和位置确定单元 313b。
定位信息获取单元 311b,用于获取用户集合中用户所对应的定位信息。
本实施例中, 定位信息获取单元 311b获取由 GPS定位得到的或者用户输入 的定位信息, 进而根据定位信息即可获知用户当前所在的位置。
位置确定单元 313b,用于根据定位信息中的位置确定用户对应的位置范围 对用户对应的位置范围取并集得到地点范围设定信息。
本实施例中,位置确定单元 313b获取用户所采用的交通方式或者当前速度, 以根据用户当前所采用的交通方式或者当前速度得到该用户对应的半径, 进而 根据定位信息中的位置和半径即可得到相应的圆, 该圆即为用户对应的位置范 围, 此时, 将用户集合中所有用户的圆进行求并就得到了地点范围设定信息中 所有用户的地点范围。
第二划分单元 330b,用于将地点范围设定信息中的地点范围划分为若干个 地点子范围。
本实施例中, 第二划分单元 330b划分地点范围为多个部分, 以得到若干个 地点子范围。
如图 16所示,在一个实施例中,上述第二划分单元 330b包括布设单元 331b 和网格划分单元 333b。
布设单元 331b,用于根据地点范围设定信息中的地点范围布设图示。
本实施例中, 布设单元 331b使地点范围以图示的形式存在, 以便于更为直 观地对地点范围进行划分。
网格划分单元 333b,用于将图示划分为若干个网格, 该网格即为地点子范 围。
本实施例中,网格划分单元 333b对图示进行网格划分, 以得到 m' n的网格, 每一网格即为一个地点子范围。
第二子区间统计单元 350b,用于对用户集合中的用户统计每一地点子范围 所对应的用户可接受次数, 并将用户可接受次数最大的地点子范围作为用户集 合对应的地点期望范围。
本实施例中, 第二子区间统计单元 350b使用长度为 m' n的计数器 C对地点 子范围进行计数, 所得到的数组中元素 即为第 i个地点子范围的用户可接受 次数, 数值最大的元素所对应的地点子范围即为用户集合对应的地点期望范围。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程, 是可以通过计算机程序来指令相关的硬件来完成, 所述的程序可存储于一计算 机可读取存储介质中, 该程序在执行时, 可包括如上述各方法的实施例的流程。 其中, 所述的存储介质可为磁碟、 光盘、 只读存储记忆体 (Read-Only Memory, ROM)或随机存储记忆体 ( Random Access Memory, RAM)等。
以上所述实施例仅表达了本发明的几种实施方式, 其描述较为具体和详细, 但并不能因此而理解为对本发明专利范围的限制。 应当指出的是, 对于本领域 的普通技术人员来说, 在不脱离本发明构思的前提下, 还可以做出若干变形和 改进, 这些都属于本发明的保护范围。 因此, 本发明专利的保护范围应以所附 权利要求为准。

Claims

权 利 要 求 书
1、 一种实现用户信息聚类的方法, 包括如下步骤:
通过量化用户信息得到相应的用户特征, 聚类所述用户特征得到用户集合; 获取所述用户集合的范围设定信息, 对所述范围设定信息进行统计以得到 所述用户集合的期望范围;
根据所述用户集合和所述期望范围生成聚类结果。
2、 根据权利要求 1所述的方法, 其特征在于, 所述通过量化用户信息得到 相应的用户特征, 聚类所述用户特征得到用户集合的步骤包括:
根据所述用户特征构造谱空间, 并进行拉普拉斯特征映射得到用户在谱空 间的顶点, 通过所述用户在谱空间的顶点对用户进行聚类得到用户集合。
3、 根据权利要求 2所述的方法, 其特征在于, 所述根据用户信息构造谱空 间, 并进行拉普拉斯特征映射得到用户在谱空间的顶点, 通过所述用户在谱空 间的顶点对用户进行聚类得到用户集合的步骤包括:
量化用户信息以构造相似矩阵;
由所述相似矩阵提取拉普拉斯矩阵, 对所述拉普拉斯矩阵进行特征分解以 构造谱空间;
将量化的用户信息映射至谱空间得到用户在谱空间的顶点;
将谱空间中距离上相互集中的顶点所对应的用户形成用户集合。
4、 根据权利要求 3所述的方法, 其特征在于, 所述量化用户信息以构造相 似矩阵的步骤包括:
根据用户信息构建每一用户的向量;
对所述向量中的元素进行归一化处理, 并将归一化处理后的元素进行加权 计算得到所述用户的量化值;
计算所述用户的量化值之间的距离得到两个用户之间的相似度, 并通 过所 述两个用户之间的相似度形成邻接矩阵;
通过所述邻接矩阵得到相似矩阵。
5、 根据权利要求 1所述的方法, 其特征在于, 所述范围设定信息为时间范 围设定信息; 所述获取所述用户的范围设定信息, 对所述范围设定信息进行统 计以得到用户集合的期望范围的步骤包括: 获取所述用户的时间范围设定信息;
将所述时间范围设定信息中的时间范围划分为若干个时间子区间; 对所述用户集合中的用户统计每一时间子区间的用户可接受次数, 并将用 户可接受次数最大的时间子区间作为所述用户集合的时间期望范围。
6、 根据权利要求 1所述的方法, 其特征在于, 所述范围设定信息为地点范 围设定信息; 所述获取所述用户的范围设定信息, 对所述范围设定信息进行统 计以得到用户集合的期望范围的步骤包括:
获取所述用户的地点范围设定信息;
将所述地点范围设定信息中的地点范围划分为若干个地点子范围; 对所述用户集合中的用户统计每一地点子范围的用户可接受次数, 并将用 户可接受次数最大的地点子范围作为所述用户集合的地点期望范 围。
7、 根据权利要求 6所述的方法, 其特征在于, 所述获取所述用户的地点范 围设定信息的步骤包括:
获取用户集合中用户的定位信息;
根据所述定位信息中的位置确定所述用户的位置范围, 对所述用户的位置 范围取并集得到所述地点范围设定信息。
8、 根据权利要求 6所述的方法, 其特征在于, 所述将所述地点范围设定信 息中的地点范围划分为若干个地点子范围的步骤包括:
根据所述地点范围设定信息中的地点范围布设图示;
将所述图示划分为若干个网格, 其中所述网格为地点子范围。
9、 一种实现用户信息聚类的装置, 其特征在于, 包括:
聚类模块, 用于量化用户信息得到相应的用户特征, 聚类所述用户特征得 到用户集合;
范围统计模块, 用于获取所述用户集合的范围设定信息, 对所述范围设定 信息进行统计以得到用户集合的期望范围;
结果生成模块, 用于根据所述用户集合和所述期望范围生成聚类结果。
10、 根据权利要求 9所述的装置, 其特征在于, 所述聚类模块还用于根据 用户信息构造谱空间, 并进行拉普拉斯特征映射得到用户在谱空间的顶点, 通 过所述用户在谱空间的顶点对用户进行聚类得到用户集合。
11、 根据权利要求 9所述的装置, 其特征在于, 所述聚类模块包括: 量化单元, 用于量化用户信息以构造相似矩阵;
谱空间构造单元, 用于由所述相似矩阵提取拉普拉斯矩阵, 对所述拉普拉 斯矩阵进行特征分解以构造谱空间;
映射单元, 用于将量化的用户信息映射至谱空间得到用户在谱空间的顶点; 集合形成单元, 用于将谱空间中距离上相互集中的顶点所对应的用户形成 用户集合。
12、 根据权利要求 11所述的装置, 其特征在于, 所述量化单元包括: 向量构建单元, 用于根据用户信息构建每一用户的向量;
运算单元, 用于对所述向量中的元素进行归一化处理, 并将归一化处理后 的元素进行加权计算得到所述用户的量化值;
相似性计算单元, 用于计算所述用户的量化值之间的距离得到两个用 户之 间的相似度, 并通过所述两个用户之间的相似度形成邻接矩阵;
相似矩阵获取单元, 用于通过所述邻接矩阵得到相似矩阵。
13、 根据权利要求 9所述的装置, 其特征在于, 所述范围设定信息为时间 范围设定信息; 所述范围统计模块包括:
第一信息获取单元, 用于获取所述用户集合的时间范围设定信息; 第一划分单元, 用于将所述时间范围设定信息中的时间范围划分为若干个 时间子区间;
第一子区间统计单元, 用于对所述用户集合中的用户统计每一时间子区间 的用户可接受次数, 并将用户可接受次数最大的时间子区间作为所述用 户集合 的时间期望范围。
14、 根据权利要求 9所述的装置, 其特征在于, 所述范围设定信息为地点 范围设定信息; 所述范围统计模块包括:
第二信息获取单元, 用于获取所述用户集合的地点范围设定信息; 第二划分单元, 用于将所述地点范围设定信息中的地点范围划分为若干个 地点子范围;
第二子区间统计单元, 用于对所述用户集合中的用户统计每一地点子范围 的用户可接受次数, 并将用户可接受次数最大的地点子范围作为用户集 合的地 点期望范围。
15、 根据权利要求 14所述的装置, 其特征在于, 所述第二信息获取单元包 括:
定位信息获取单元, 用于获取用户集合中用户的定位信息;
位置确定单元, 用于根据所述定位信息中的位置确定所述用户的位置 范 围, 对所述用户的位置范围取并集得到所述地点范围设定信息。
16、 根据权利要求 14所述的装置, 其特征在于, 所述第二划分单元包括: 布设单元, 用于根据所述地点范围设定信息中的地点范围布设图示; 网格划分单元, 用于将所述图示划分为若干个网格, 其中所述网格为地点 子范围。
17、一个或多个计算机可读存取介质,包含用于执行根据权利要求 1-8中任 何一个所述方法的计算机可执行程序。
18、 一种装置, 包括:
处理器; 以及
存储器, 含有计算机可执行程序, 所述存储器和计算机可执行程序被配置 为利用所述处理器使得所述装置:
通过量化用户信息得到相应的用户特征, 聚类所述用户特征得到用户集合; 获取所述用户集合的范围设定信息, 对所述范围设定信息进行统计以得到 所述用户集合的期望范围;
根据所述用户集合和所述期望范围生成聚类结果。
PCT/CN2014/084484 2013-08-26 2014-08-15 实现用户信息聚类的方法和装置 WO2015027828A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201310376923.2A CN104424235B (zh) 2013-08-26 2013-08-26 实现用户信息聚类的方法和装置
CN201310376923.2 2013-08-26

Publications (1)

Publication Number Publication Date
WO2015027828A1 true WO2015027828A1 (zh) 2015-03-05

Family

ID=52585547

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2014/084484 WO2015027828A1 (zh) 2013-08-26 2014-08-15 实现用户信息聚类的方法和装置

Country Status (2)

Country Link
CN (1) CN104424235B (zh)
WO (1) WO2015027828A1 (zh)

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10419466B2 (en) * 2016-02-09 2019-09-17 Darktrace Limited Cyber security using a model of normal behavior for a group of entities
US10516693B2 (en) 2016-02-25 2019-12-24 Darktrace Limited Cyber security
US10986121B2 (en) 2019-01-24 2021-04-20 Darktrace Limited Multivariate network structure anomaly detector
US11075932B2 (en) 2018-02-20 2021-07-27 Darktrace Holdings Limited Appliance extension for remote communication with a cyber security appliance
US11463457B2 (en) 2018-02-20 2022-10-04 Darktrace Holdings Limited Artificial intelligence (AI) based cyber threat analyst to support a cyber security appliance
US11470103B2 (en) 2016-02-09 2022-10-11 Darktrace Holdings Limited Anomaly alert system for cyber threat detection
US11477222B2 (en) 2018-02-20 2022-10-18 Darktrace Holdings Limited Cyber threat defense system protecting email networks with machine learning models using a range of metadata from observed email communications
US11693964B2 (en) 2014-08-04 2023-07-04 Darktrace Holdings Limited Cyber security using one or more models trained on a normal behavior
US11709944B2 (en) 2019-08-29 2023-07-25 Darktrace Holdings Limited Intelligent adversary simulator
US11924238B2 (en) 2018-02-20 2024-03-05 Darktrace Holdings Limited Cyber threat defense system, components, and a method for using artificial intelligence models trained on a normal pattern of life for systems with unusual data sources
US11936667B2 (en) 2020-02-28 2024-03-19 Darktrace Holdings Limited Cyber security system applying network sequence prediction using transformers
US11962552B2 (en) 2018-02-20 2024-04-16 Darktrace Holdings Limited Endpoint agent extension of a machine learning cyber defense system for email
US11973774B2 (en) 2020-02-28 2024-04-30 Darktrace Holdings Limited Multi-stage anomaly detection for process chains in multi-host environments
US11985142B2 (en) 2020-02-28 2024-05-14 Darktrace Holdings Limited Method and system for determining and acting on a structured document cyber threat risk
US12034767B2 (en) 2019-08-29 2024-07-09 Darktrace Holdings Limited Artificial intelligence adversary red team
US12063243B2 (en) 2018-02-20 2024-08-13 Darktrace Holdings Limited Autonomous email report generator

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106484723B (zh) * 2015-08-28 2019-08-02 杭州华为数字技术有限公司 Egonet的信息处理方法和装置
CN106506705B (zh) * 2016-12-29 2020-07-28 平安科技(深圳)有限公司 基于位置服务的人群分类方法及装置
CN109062914B (zh) * 2017-06-12 2020-10-23 东软集团股份有限公司 用户推荐方法和装置,存储介质和服务器
CN108599140B (zh) * 2018-01-24 2021-01-29 合肥工业大学 用电负荷特征分析方法和装置、存储介质
CN110826605A (zh) * 2019-10-24 2020-02-21 北京明略软件系统有限公司 一种跨平台识别用户的方法及装置
CN111060124A (zh) * 2019-12-18 2020-04-24 深圳集智数字科技有限公司 一种确定聚集地点的方法和相关装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101996213A (zh) * 2009-08-27 2011-03-30 顾海松 一种实时客户购物行为分析系统及其方法
CN102254028A (zh) * 2011-07-22 2011-11-23 青岛理工大学 一种集成属性和结构相似性的个性化商品推荐方法和系统
CN102591872A (zh) * 2011-01-06 2012-07-18 中国移动通信集团四川有限公司 一种客户特征库生成方法和装置
CN102654860A (zh) * 2011-03-01 2012-09-05 北京彩云在线技术开发有限公司 一种个性化音乐推荐方法及系统

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070011224A1 (en) * 1999-10-22 2007-01-11 Jesus Mena Real-time Internet data mining system and method for aggregating, routing, enhancing, preparing, and analyzing web databases
CN101420313B (zh) * 2007-10-22 2011-01-12 北京搜狗科技发展有限公司 一种针对客户端用户群进行聚类的方法和系统
CN102506884A (zh) * 2011-10-28 2012-06-20 百度在线网络技术(北京)有限公司 基于地图为多个用户推荐聚会地点的方法、系统及装置
CN102594905B (zh) * 2012-03-07 2014-07-16 南京邮电大学 一种基于场景的社交网络位置兴趣点推荐方法
CN103077220B (zh) * 2012-12-29 2016-06-29 中国科学院深圳先进技术研究院 一种基于用户群关联度的个性化推荐方法及系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101996213A (zh) * 2009-08-27 2011-03-30 顾海松 一种实时客户购物行为分析系统及其方法
CN102591872A (zh) * 2011-01-06 2012-07-18 中国移动通信集团四川有限公司 一种客户特征库生成方法和装置
CN102654860A (zh) * 2011-03-01 2012-09-05 北京彩云在线技术开发有限公司 一种个性化音乐推荐方法及系统
CN102254028A (zh) * 2011-07-22 2011-11-23 青岛理工大学 一种集成属性和结构相似性的个性化商品推荐方法和系统

Cited By (33)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11693964B2 (en) 2014-08-04 2023-07-04 Darktrace Holdings Limited Cyber security using one or more models trained on a normal behavior
US11470103B2 (en) 2016-02-09 2022-10-11 Darktrace Holdings Limited Anomaly alert system for cyber threat detection
US10419466B2 (en) * 2016-02-09 2019-09-17 Darktrace Limited Cyber security using a model of normal behavior for a group of entities
US10516693B2 (en) 2016-02-25 2019-12-24 Darktrace Limited Cyber security
US11689557B2 (en) 2018-02-20 2023-06-27 Darktrace Holdings Limited Autonomous report composer
US11962552B2 (en) 2018-02-20 2024-04-16 Darktrace Holdings Limited Endpoint agent extension of a machine learning cyber defense system for email
US11418523B2 (en) 2018-02-20 2022-08-16 Darktrace Holdings Limited Artificial intelligence privacy protection for cybersecurity analysis
US11457030B2 (en) 2018-02-20 2022-09-27 Darktrace Holdings Limited Artificial intelligence researcher assistant for cybersecurity analysis
US11463457B2 (en) 2018-02-20 2022-10-04 Darktrace Holdings Limited Artificial intelligence (AI) based cyber threat analyst to support a cyber security appliance
US11336670B2 (en) 2018-02-20 2022-05-17 Darktrace Holdings Limited Secure communication platform for a cybersecurity system
US11477222B2 (en) 2018-02-20 2022-10-18 Darktrace Holdings Limited Cyber threat defense system protecting email networks with machine learning models using a range of metadata from observed email communications
US11477219B2 (en) 2018-02-20 2022-10-18 Darktrace Holdings Limited Endpoint agent and system
US11522887B2 (en) 2018-02-20 2022-12-06 Darktrace Holdings Limited Artificial intelligence controller orchestrating network components for a cyber threat defense
US11546360B2 (en) 2018-02-20 2023-01-03 Darktrace Holdings Limited Cyber security appliance for a cloud infrastructure
US11546359B2 (en) 2018-02-20 2023-01-03 Darktrace Holdings Limited Multidimensional clustering analysis and visualizing that clustered analysis on a user interface
US11606373B2 (en) 2018-02-20 2023-03-14 Darktrace Holdings Limited Cyber threat defense system protecting email networks with machine learning models
US11075932B2 (en) 2018-02-20 2021-07-27 Darktrace Holdings Limited Appliance extension for remote communication with a cyber security appliance
US11689556B2 (en) 2018-02-20 2023-06-27 Darktrace Holdings Limited Incorporating software-as-a-service data into a cyber threat defense system
US12063243B2 (en) 2018-02-20 2024-08-13 Darktrace Holdings Limited Autonomous email report generator
US11336669B2 (en) 2018-02-20 2022-05-17 Darktrace Holdings Limited Artificial intelligence cyber security analyst
US11716347B2 (en) 2018-02-20 2023-08-01 Darktrace Holdings Limited Malicious site detection for a cyber threat response system
US11799898B2 (en) 2018-02-20 2023-10-24 Darktrace Holdings Limited Method for sharing cybersecurity threat analysis and defensive measures amongst a community
US11843628B2 (en) 2018-02-20 2023-12-12 Darktrace Holdings Limited Cyber security appliance for an operational technology network
US11902321B2 (en) 2018-02-20 2024-02-13 Darktrace Holdings Limited Secure communication platform for a cybersecurity system
US11924238B2 (en) 2018-02-20 2024-03-05 Darktrace Holdings Limited Cyber threat defense system, components, and a method for using artificial intelligence models trained on a normal pattern of life for systems with unusual data sources
US10986121B2 (en) 2019-01-24 2021-04-20 Darktrace Limited Multivariate network structure anomaly detector
US11709944B2 (en) 2019-08-29 2023-07-25 Darktrace Holdings Limited Intelligent adversary simulator
US12034767B2 (en) 2019-08-29 2024-07-09 Darktrace Holdings Limited Artificial intelligence adversary red team
US11936667B2 (en) 2020-02-28 2024-03-19 Darktrace Holdings Limited Cyber security system applying network sequence prediction using transformers
US11973774B2 (en) 2020-02-28 2024-04-30 Darktrace Holdings Limited Multi-stage anomaly detection for process chains in multi-host environments
US11985142B2 (en) 2020-02-28 2024-05-14 Darktrace Holdings Limited Method and system for determining and acting on a structured document cyber threat risk
US11997113B2 (en) 2020-02-28 2024-05-28 Darktrace Holdings Limited Treating data flows differently based on level of interest
US12069073B2 (en) 2020-02-28 2024-08-20 Darktrace Holdings Limited Cyber threat defense system and method

Also Published As

Publication number Publication date
CN104424235B (zh) 2018-01-05
CN104424235A (zh) 2015-03-18

Similar Documents

Publication Publication Date Title
WO2015027828A1 (zh) 实现用户信息聚类的方法和装置
WO2018153292A1 (zh) 社交圈子的生成方法、装置及存储介质
US9510141B2 (en) App recommendation using crowd-sourced localized app usage data
WO2018054342A1 (zh) 一种网络数据流分类的方法及系统
CN109167835A (zh) 一种基于kubernetes的物理资源调度方法及系统
WO2019134274A1 (zh) 兴趣探索方法、存储介质、电子设备及系统
US20110213785A1 (en) Social Data Ranking and Processing
US20170068902A1 (en) Modeling of Geospatial Location Over Time
WO2022001918A1 (zh) 构建预测模型的方法、装置、计算设备和存储介质
JP2016503937A (ja) 活動領域の検索および決定
CN103699678B (zh) 一种基于多阶段分层采样的层次聚类方法和系统
WO2011106950A1 (zh) 复杂网络配置方法和系统及服务器资源配置管理模块
CN115004210A (zh) 用户画像列表构建方法、装置、服务器以及存储介质
MX2014004804A (es) Analisis periodico de forma de onda ambiental, para una configuracion dinamica del dispositivo.
CN107526975A (zh) 一种基于差分隐私保护决策树的方法
Zhang et al. Encrypted traffic classification based on an improved clustering algorithm
CN110298687B (zh) 一种区域吸引力评估方法及设备
CN104572766B (zh) 一种社交网络的用户状态识别方法及装置
CN113379176A (zh) 电信网络异常数据检测方法、装置、设备和可读存储介质
WO2019062419A1 (zh) 应用清理方法、装置、存储介质及电子设备
CN105335476B (zh) 一种热点事件分类方法及装置
Majumdar et al. Energy efficient e-healthcare framework using HWPSO-based clustering approach
Aldhyani et al. Intelligent time series model to predict bandwidth utilization
Takyi et al. Clustering techniques for traffic classification: a comprehensive review
CN112214684B (zh) 一种种子扩展的重叠社区发现方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 14839287

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 160616)

122 Ep: pct application non-entry in european phase

Ref document number: 14839287

Country of ref document: EP

Kind code of ref document: A1