Disclosure of Invention
The embodiment of the invention provides an advertisement putting effect evaluation method, an advertisement putting effect evaluation device and electronic equipment.
In order to achieve the above purpose, the embodiment of the invention adopts the following technical scheme:
in a first aspect, an embodiment of the present invention provides an advertisement delivery effect evaluation method, where the method includes: acquiring feedback information corresponding to each advertisement in real time;
distributing the feedback information to one of a plurality of preset local clustering modules based on a set rule;
clustering operation is carried out on the feedback information through the local clustering module to obtain a first set number of clustering centers;
sending a plurality of clustering centers obtained by the operation of each preset local clustering module to a global clustering module in real time;
performing clustering operation on the received multiple clustering centers through the global clustering module to obtain a second set number of target clustering centers;
evaluating the delivery effect of each delivered advertisement according to the second set number of target clustering centers and the feedback information;
the number of the preset local clustering modules is at least two, and the feedback information comprises attribute information of the delivered advertisements and demand information of the users for the delivered advertisements.
Further, the allocating the feedback information to one of the preset local clustering modules based on the setting rule includes:
time numbering is carried out on the feedback information corresponding to each advertisement put through a feedback information distribution module based on the time sequence;
respectively carrying out remainder taking operation on the time number corresponding to each piece of feedback information and the total number of the preset local clustering modules to obtain a remainder taking result;
and distributing the feedback information to a local clustering module with the sequence number of the surplus result.
Further, the performing clustering operation on the received multiple clustering centers through the global clustering module to obtain a second set number of target clustering centers includes:
and performing clustering operation on the received multiple clustering centers through the global clustering module based on a K-means + + clustering algorithm to obtain a second set number of target clustering centers.
Further, the evaluating the delivery effect of each delivered advertisement according to the second set number of target cluster centers and the feedback information includes:
respectively calculating the number of the feedback information in the cluster corresponding to each target cluster center;
and evaluating the delivery effect of each delivered advertisement according to the data volume of the feedback information in each cluster.
Further, after the global clustering module performs clustering operation on the received multiple clustering centers to obtain a second set number of target clustering centers, the method further includes:
and sending the second set number of target clustering centers to a priori knowledge storage module so that the priori knowledge stored in the priori knowledge storage module is updated by the priori knowledge storage module.
Further, before the local clustering module performs clustering operation on the feedback information to obtain a first set number of clustering centers, the method further includes:
acquiring prior knowledge from the prior knowledge storage module;
initializing the local clustering module based on the prior knowledge.
Further, the method also comprises the following steps: and performing data compression processing on the elements in each class cluster in a characteristic vector mode.
Further, the performing data compression processing on the elements in each class cluster by means of a feature vector includes:
and performing data compression processing on the elements in each class cluster according to the following formula:
wherein,
a feature vector representing the cluster c of the class,
represents the sum of squares of the feature values of each dimension of each element in the class cluster c,
representing the sum of the characteristic values of the dimensions of the elements of the class c, CF2
tCF1, which represents the sum of squares of the time numbers corresponding to the elements in the class c
tAnd n is the number of elements in the class cluster c.
Further, the performing clustering operation on the feedback information through the local clustering module to obtain a first set number of clustering centers includes:
respectively calculating the distance between the feedback information and the clustering centers corresponding to various clusters in the local clustering module;
if the distance between the feedback information and the clustering center corresponding to one of the various clusters is smaller than the neighborhood radius of the one of the various clusters, the feedback information is merged into the one of the various clusters, and the current various clusters in the local clustering module are operated based on a set clustering algorithm to obtain a first set number of clustering centers;
and if the distance between the feedback information and the clustering centers corresponding to the various clusters is not less than the neighborhood radius corresponding to the various clusters, merging or deleting the various clusters existing in the local clustering module, creating a new cluster by taking the feedback information as an element, and calculating the current various clusters in the local clustering module based on a set clustering algorithm to obtain the clustering centers of a first set number.
Further, the neighborhood radius corresponding to each cluster is:
wherein r (c) represents the neighborhood radius corresponding to the class cluster c, other represents the class cluster closest to the class cluster c, n
cDenotes the number of elements in the class cluster c, n
otherIndicates the number of elements in the class cluster other,
representing the root mean square value of the elements in the class cluster c,
represents the average value of the elements in the class cluster c,
to represent
And
the euclidean distance between them,
represents the average value of the elements in the cluster-like other,
to represent
And
the euclidean distance between.
Further, the merging or deleting operations of the various existing clusters in the local clustering module include:
determining the activity degree of each type of cluster based on the number of elements in each type of cluster existing in the local clustering module and the old and new degree of the elements;
if the cluster with the activity degree lower than the set value exists, deleting the cluster with the activity degree lower than the set value;
and if the cluster with the activity degree lower than the set value does not exist, combining the two closest clusters into one cluster.
Further, the determining the activity degree of each type of cluster based on the number of elements in each type of cluster existing in the local clustering module and the freshness degree of the elements includes:
counting the merging time of the latest elements with the set number in each type of clusters;
calculating the average merging time of the latest elements with set quantity in each type of clusters;
and determining the average merging time of the latest set number of elements in each type of cluster as the activity degree of each type of cluster.
Further, the merging two closest class clusters into one class cluster includes:
adding the feature vectors respectively corresponding to the two cluster classes with the shortest distance;
and taking the added feature vector as the feature vector of the new cluster after the two closest cluster classes are merged.
In a second aspect, an embodiment of the present invention provides an advertisement delivery effect evaluation apparatus, where the apparatus includes:
the acquisition module is used for acquiring feedback information corresponding to each advertisement put in real time;
the distribution module is used for distributing the feedback information to one of a plurality of preset local clustering modules based on a set rule;
the first clustering module is used for performing clustering operation on the feedback information through the local clustering module to obtain a first set number of clustering centers;
the sending module is used for sending a plurality of clustering centers obtained by the operation of the preset local clustering modules to the global clustering module in real time;
the second clustering module is used for carrying out clustering operation on the received multiple clustering centers through the global clustering module to obtain a second set number of target clustering centers;
the evaluation module is used for evaluating the delivery effect of each delivered advertisement according to the second set number of target clustering centers and the feedback information;
the number of the preset local clustering modules is at least two, and the feedback information comprises attribute information of the delivered advertisements and demand information of the users for the delivered advertisements.
In a third aspect, an embodiment of the present invention provides an electronic device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor, when executing the computer program, implements the advertisement impression effect evaluation method according to the first aspect.
In a fourth aspect, embodiments of the present invention provide a storage medium containing computer-executable instructions, which when executed by a computer processor, implement the advertisement placement effectiveness evaluation method according to the first aspect.
According to the advertisement putting effect evaluation method provided by the embodiment of the invention, distributed clustering operation is carried out on feedback information of the put advertisements through at least two local clustering modules respectively, then clustering is carried out on clustering centers obtained by the operation of the distributed local clustering modules through a global clustering module again to finally obtain a target clustering center, and the putting effect of the put advertisements is evaluated according to the target clustering center and the feedback information of the put advertisements, so that the purpose of evaluating the putting effect of the advertisements in real time is realized, and then advertisement operators are helped to adjust putting strategies in time to realize the maximization of putting benefits, and meanwhile, the problem of low single-machine clustering efficiency is solved through a distributed clustering mode.
Detailed Description
In order to make the technical problems solved, technical solutions adopted and technical effects achieved by the present invention clearer, the technical solutions of the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example one
Fig. 1 is a schematic flow chart of an advertisement delivery effect evaluation method according to an embodiment of the present invention. The advertisement effectiveness evaluation method disclosed in this embodiment may be executed by an advertisement effectiveness evaluation device, where the device may be implemented by software and/or hardware and is generally integrated in a terminal, such as a server. Referring specifically to fig. 1, the method comprises the steps of:
110. and acquiring feedback information corresponding to each advertisement in real time.
The feedback information comprises attribute information of the delivered advertisements and demand information of the users for the delivered advertisements. The attribute information may specifically refer to the type of advertisement delivered (such as an electric appliance, a book, etc.); the advertisement delivery position specifically refers to the position of the advertisement on a delivery website; the loading speed of the served advertisements, the source of the served advertisements (e.g., Taobao, Jingdong, etc.), and the like. The information of the user's demand for advertisement delivery specifically indicates whether the user clicks to view the delivered advertisement, if the user clicks to view the delivered advertisement, it indicates that the user has a demand for the advertisement, and if many users click to view the advertisement, it indicates that the advertisement delivery effect is good, and the advertisement delivery is successful.
Specifically, the feedback information corresponding to each advertisement delivered may be represented by the following format: the meaning of each field of info ═ ui, ad _ type, ad _ pos, type, speed, source, gif, size) is shown in table 1 below;
the exposure indicates whether the user sees the delivered advertisement, the mark click indicates whether the user clicks the delivered advertisement, and the information of each field of the delivered advertisement shown in the table 1 is fed back by the background server in real time according to the behavior of the user.
120. And distributing the feedback information to one of a plurality of preset local clustering modules based on a set rule.
Specifically, the set rule may be that feedback information corresponding to each advertisement delivered is evenly distributed to each preset local clustering module, so that each local clustering module has the same load, each local clustering module has the same clustering speed as much as possible, and the real-time performance of advertisement delivery effect evaluation is improved.
Illustratively, the feedback information corresponding to each delivered advertisement can be averagely distributed to each preset local clustering module in the following way:
time numbering is carried out on the feedback information corresponding to each advertisement put through a feedback information distribution module based on the time sequence;
respectively carrying out remainder taking operation on the time number corresponding to each piece of feedback information and the total number of the preset local clustering modules to obtain a remainder taking result;
and distributing the feedback information to a local clustering module with the sequence number of the surplus result.
The feedback information corresponding to each advertisement delivered by the background is passed through the feedback information distribution module, and the feedback information distribution module time-numbers the feedback information corresponding to each advertisement delivered based on the time sequence, for example, a counter may be set in the feedback information distribution module, counting is started from 0, the current count value is assigned to the feedback information for each feedback information, so as to serve as the time number of the feedback information, and the count value of the counter is increased by 1. When initializing each local clustering module, setting a corresponding serial number for each local clustering module, in this embodiment, taking three local clustering modules as an example, where the corresponding serial numbers are 0, 1, and 2, respectively, and performing a remainder operation on the time number corresponding to each piece of feedback information and the total number of preset local clustering modules to obtain a remainder result, for example, assuming that the time number of a certain piece of feedback information is 5, performing a remainder operation on 5 and 3, and setting the remainder result to be 2, allocating the feedback information to the local clustering module with the serial number of 2.
The number of the local clustering modules can be expanded and is not limited to 3, and the data volume of the delivered advertisements is large, so that the data volume of the feedback information corresponding to each delivered advertisement is also large, and the clustering operation of the feedback information corresponding to each delivered advertisement is performed by arranging a plurality of distributed local clustering modules together, so that the clustering speed and the processing capacity can be improved, and the real-time performance of the advertisement delivery effect evaluation can be improved.
Furthermore, before the feedback information is distributed to the preset local clustering modules, the preset local clustering modules are initialized by using the priori knowledge, so that the local clustering modules have a certain number of clusters, and the speed of clustering the new feedback information is increased. The priori knowledge is usually a set number of clustering centers obtained by clustering a certain number of feedback information in advance, and the priori knowledge is used for initializing each preset local clustering module, and specifically comprises the following steps: and storing the preset number of clustering centers to each local clustering module.
130. And performing clustering operation on the feedback information through the local clustering module to obtain a first set number of clustering centers.
Specifically, the clustering operation performed on the feedback information by the local clustering module is as follows: and integrating new feedback information into a cluster corresponding to an existing cluster center in the local clustering module through a specific clustering algorithm, updating the cluster center of the current cluster, or building a new cluster by taking the new feedback information as the cluster center, and deleting or combining the existing clusters in the local clustering module based on a certain rule to ensure that the number of the cluster centers is unchanged.
The specific clustering algorithm can be a DBSCAN algorithm, the DBSCAN algorithm is an excellent clustering algorithm based on density, the number of classified points is not required to be preset, noise points can be effectively found, the time complexity is high, the clustering speed is rapidly reduced under the condition of huge data volume, and the large data volume is subjected to distributed dispersion in a distributed local clustering mode to solve the problem of rapid reduction of the clustering speed.
140. And sending a plurality of clustering centers obtained by the operation of each preset local clustering module to the global clustering module in real time.
The preset local clustering modules work in parallel, clustering operation is carried out on the distributed feedback information corresponding to the advertisements, and the clustering centers obtained by the layout clustering modules are sent to the global clustering module in real time.
150. And performing clustering operation on the received multiple clustering centers through the global clustering module to obtain a second set number of target clustering centers.
Specifically, the data amount of the cluster center to be processed by the global clustering module is much smaller than the data amount of the feedback information processed by the local clustering module, so that the global clustering module can process the plurality of cluster centers by using an algorithm with higher clustering quality. Since the conventional clustering algorithm needs to perform full-disk scanning on a data set to be clustered more than once, a task mechanism can be adopted in the global clustering module. The task mechanism specifically means that when each batch of clustering centers (from clustering centers of a plurality of local clustering modules) arrives, a transaction is created to temporarily store the clustering centers, operate a clustering algorithm and record clustering results, and the three processes are automatically closed after being executed. In order to flexibly control the number of clusters and not need prior knowledge of data characteristics, the global clustering module can adopt a clustering algorithm K-means + + based on division to perform clustering operation on the received multiple clustering centers to obtain a second set number of target clustering centers.
The flow of the K-means + + clustering algorithm is as follows:
inputting: k-number of target clustering centers, D-data set of clustering centers containing n local clustering modules
And (3) outputting: set of k target cluster centers
1) Selecting k objects from D as initial clustering centers;
2) calculating the distance between each object and the k initial clustering centers according to the mean value of each object in the D, and dividing each object in the D into corresponding clusters corresponding to the initial clustering centers according to the minimum distance;
3) recalculating the mean value of the cluster class with each element changed (namely the cluster center);
4) and (5) circularly executing the steps 2) -3) until the cluster center is not changed any more, and finishing the clustering.
The process of selecting k objects from D as the initial clustering centers in step 1) may specifically be:
setting seed point seeds to be 0;
randomly selecting an object from the D as a first clustering center;
seeds++;
if seeds < k, calculating the distance D (x) between all the objects x which are not selected as the clustering center in the data set D and the objects selected as the clustering center;
selecting an object as a clustering center, wherein the selection principle is as follows: the larger (x) is, the higher is the probability that the object x is selected as the cluster center;
seeds++;
if seeds is equal to k, the process ends.
160. And evaluating the delivery effect of each delivered advertisement according to the second set number of target clustering centers and the feedback information.
Specifically, the number of the feedback information in the cluster corresponding to each target cluster center is respectively calculated;
and evaluating the delivery effect of each delivered advertisement according to the data volume of the feedback information in each cluster.
When the target clustering centers are obtained, the number of the feedback information in the clusters corresponding to each target clustering center is respectively calculated, and the cluster with more elements shows that the advertisement delivery effect corresponding to the feedback information in the cluster is better.
According to the advertisement putting effect evaluation method provided by the embodiment, firstly, the distributed local clustering modules are used for respectively carrying out clustering operation on feedback information of the put advertisements, then the global clustering module is used for clustering the clustering centers obtained by the operation of the distributed local clustering modules again to finally obtain the target clustering center, the putting effect of the put advertisements is evaluated according to the target clustering center and the feedback information of the put advertisements, the purpose of carrying out real-time evaluation on the putting effect of the advertisements is achieved, and then the advertisement operators are helped to adjust the putting strategy in time to achieve the maximization of putting benefits, and meanwhile, the problem of low single-machine clustering efficiency is solved through the distributed clustering mode.
Example two
Fig. 2 is a schematic flow chart of an advertisement delivery effect evaluation method according to a second embodiment of the present invention. On the basis of the above embodiment, the present embodiment is optimized, and further, a target clustering center output by the global clustering module is fed back to the priori knowledge storage module, so that the priori knowledge stored by the priori knowledge storage module is updated by the priori knowledge storage module, and the latest priori knowledge is obtained from the priori knowledge storage module before the feedback information is clustered by the local clustering module each time, so that the optimization has the advantages of balancing the state of the whole evaluation system, and further improving the speed, accuracy and real-time performance of online clustering by the local clustering module. Referring specifically to fig. 2, the method includes the steps of:
210. and acquiring feedback information corresponding to each advertisement in real time.
220. And distributing the feedback information to one of a plurality of preset local clustering modules based on a set rule.
230. And acquiring the prior knowledge from the prior knowledge storage module, and initializing the local clustering module based on the prior knowledge.
The prior knowledge is obtained from the prior knowledge storage module, specifically, a set number of latest clustering centers are obtained from the prior knowledge storage module, the latest clustering centers are used for replacing the existing relatively old clustering centers, and meanwhile, the existing clusters of the local module are emptied. The latest refers to the timestamp latest, for example: cluster center a has a timestamp of 10 at 2018.06.19 days: 01, score, and the timestamp of cluster center b is 10.02 scores of 2018.06.19 days, then cluster center b is the newer cluster center relative to cluster center a. The set number of the latest clustering centers is specifically as follows: assuming that a, b, c, d and e5 cluster centers are in total in the prior knowledge storage module, and the time stamp corresponding to each cluster center is 10 on the same day: 01. 10:02, 10:03, 10:04 and 10:05, then the 3 most recent cluster centers obtained from them are: and acquiring clustering centers c, d and e.
240. And performing clustering operation on the feedback information through the local clustering module to obtain a first set number of clustering centers.
250. And sending a plurality of clustering centers obtained by the operation of each preset local clustering module to the global clustering module in real time.
260. And performing clustering operation on the received multiple clustering centers through the global clustering module to obtain a second set number of target clustering centers.
270a, evaluating the delivery effect of each advertisement according to the second set number of target clustering centers and the feedback information.
270b, sending the second set number of target clustering centers to a priori knowledge storage module so that the priori knowledge stored in the priori knowledge storage module is updated by the priori knowledge storage module.
Specifically, the priori knowledge storage module compares the timestamp of the received target clustering center with the timestamp of the stored clustering centers, replaces the old clustering centers with the new clustering centers, and finally maintains the clustering centers in the priori knowledge storage module at the set number. Before the target clustering centers are fed back to the priori knowledge storage module for the first time, the priori knowledge stored in the priori knowledge storage module is the clustering centers with the set number and the corresponding clusters obtained by the traditional clustering algorithm operation according to the set number of feedback information data pre-stored on the system disk.
Referring to fig. 3, an advertisement placement effectiveness evaluation architecture diagram specifically includes: a priori knowledge storage module 310, configured to receive an initialization input and a target clustering center fed back by the global clustering module 340, and send a latest target clustering center to the local clustering module 330, so as to initialize the local clustering module 330; the feedback information distribution module 320 is configured to perform time numbering on the feedback information corresponding to each advertisement, based on a time sequence of arrival of the feedback information at the module, and evenly distribute the feedback information to the preset local clustering module 330; the local clustering module 330 is configured to perform online clustering operation on the distributed feedback information to obtain a first set number of clustering centers, and send the clustering centers to the global clustering module 340; the global clustering module 340 is configured to perform clustering operation again on the received multiple first set number of clustering centers to obtain a target clustering center, and feed back the target clustering center to the prior knowledge storage module 310 in real time while outputting the target clustering center.
In the advertisement delivery effect evaluation method provided by this embodiment, the target clustering center output by the global clustering module is fed back to the priori knowledge storage module, so that the priori knowledge stored in the priori knowledge storage module is updated by the priori knowledge storage module, and before the feedback information is clustered by the local clustering module each time, the latest priori knowledge is obtained from the priori knowledge storage module, and the initialization is performed based on the latest priori knowledge, thereby realizing the state balance of the whole evaluation system, and further improving the speed, accuracy and real-time performance of online clustering by the local clustering module.
EXAMPLE III
Fig. 4 is a schematic flow chart of an advertisement delivery effect evaluation method according to a third embodiment of the present invention. On the basis of the above embodiments, the present embodiment is further optimized. In the process of processing the feedback information data stream, high-speed data circulation and information interaction bring great pressure to a network, and in the embodiment, the data compression processing is performed on the elements in each cluster type by adopting a characteristic vector mode, so that the speed of data circulation and interaction is improved, and the data quality is ensured. Referring specifically to fig. 4, the method includes the steps of:
410. and acquiring feedback information corresponding to each advertisement in real time.
420. And distributing the feedback information to one of a plurality of preset local clustering modules based on a set rule.
430. And performing data compression processing on the elements in each class cluster in a characteristic vector mode.
Specifically, the data compression processing is performed on the feedback information in each cluster class according to the following formula:
wherein,
a feature vector representing the cluster c of the class,
represents the sum of squares of the feature values of each dimension of each element in the class cluster c,
representing the sum of the characteristic values of the dimensions of the elements of the class c, CF2
tCF1, which represents the sum of squares of the time numbers corresponding to the elements in the class c
tAnd n is the number of elements in the class cluster c.
440. And respectively calculating the distance between the feedback information and the clustering centers corresponding to various clusters in the local clustering module.
Specifically, the euclidean distance between the characteristic average value of the feedback information and the characteristic average value of each cluster is calculated to be used as the distance between the feedback information and the cluster center corresponding to each cluster.
The characteristic average value of the feedback information is
Wherein,
the sum of the characteristic values of all dimensions of the feedback information is represented, and the characteristic average value of all clusters is
Wherein,
and n is the number of elements in the class cluster c.
450a, if the distance between the feedback information and the corresponding clustering center of one of the various clusters is smaller than the neighborhood radius of the one of the various clusters, merging the feedback information into the one of the various clusters, and calculating the current various clusters in the local clustering module based on a set clustering algorithm to obtain a first set number of clustering centers.
450b, if the distance between the feedback information and the cluster centers corresponding to the various clusters is not smaller than the neighborhood radius corresponding to the various clusters, merging or deleting the various clusters existing in the local clustering module, creating a new cluster by taking the feedback information as an element, and calculating the current various clusters in the local clustering module based on a set clustering algorithm to obtain a first set number of cluster centers.
Specifically, the neighborhood radius corresponding to each cluster is:
wherein r (c) represents the neighborhood radius corresponding to the class cluster c, other represents the class cluster closest to the class cluster c, n
cDenotes the number of elements in the class cluster c, n
otherIndicates the number of elements in the class cluster other,
representing the root mean square value of the elements in the class cluster c,
represents the average value of the elements in the class cluster c,
to represent
And
the euclidean distance between them,
represents the average value of the elements in the cluster-like other,
to represent
And
the euclidean distance between.
Illustratively, the merging or deleting operation of various existing clusters in the local clustering module includes:
determining the activity degree of various clusters based on the number of elements in various clusters and the old and new degrees of the elements in the local clustering module;
if the cluster with the activity degree lower than the set value exists, deleting the cluster with the activity degree lower than the set value;
and if the cluster with the activity degree lower than the set value does not exist, combining the two closest clusters into one cluster.
When selecting the cluster to be merged or deleted, the number of elements in the cluster and the contribution of the current cluster to the whole local clustering module need to be considered, wherein the contribution is determined by the freshness of the cluster. If the number of elements of a class cluster is small but all elements are new elements, errors are inevitably caused by directly carrying out class cluster deletion operation, and therefore the freshness and the number of the elements of the class cluster need to be comprehensively considered. In some cases, an already existing cluster class accumulates a large number of elements over the history stream, but it no longer accepts new elements for a recent period of time, such cluster class is no longer active, and the contribution to the overall local clustering module should be attenuated.
In order to measure the activity degree of a class cluster, the recently accepted elements of each class cluster need to be analyzed in a statistical manner, considering the limitation of the system to the storage space, each class cluster only stores the arrival time of the latest m elements (m is far less than the number of elements in the class cluster), and the m elements adopt a storage structure of a queue to ensure the first-in first-out sequence. Calculating the average value of the time in the queue and comparing with 1/n of the current system time to judge whether the cluster is in an active state,
further, the determining the activity degree of each type of cluster based on the number of elements in each type of cluster existing in the local clustering module and the freshness degree of the elements includes:
counting the merging time of the latest elements with the set number in each type of clusters;
calculating the average merging time of the latest elements with set quantity in each type of clusters;
and determining the average merging time of the latest set number of elements in each type of cluster as the activity degree of each type of cluster.
If the sum/m > of each member in Q is equal to current/n, the cluster in which Q is located is an active cluster, and n is usually 2.
Further, the merging two closest class clusters into one class cluster includes:
adding the feature vectors respectively corresponding to the two cluster classes with the shortest distance;
and taking the added feature vector as the feature vector of the new cluster after the two closest cluster classes are merged.
Specifically, the pseudo code of the process of receiving the new feedback information by the local clustering module is as follows:
inputting:
tr-newly arrived feedback information
t _ last-a priori knowledge timestamp in current local clustering module
t _ current-time stamp stored by a priori knowledge storage module
clusterNum-cluster number of current local clustering module
And (3) outputting: clusterNum clustering centers
(1)if t_current>t_last;
(2) Initializing a current local clustering module; i.e. obtaining a set number of latest cluster centers from the prior knowledge storage module
(3) Calculating the distances from tr to the clustering centers of all the clusters, and finding the cluster c closest to the tr;
(4) distance of if tr to c center is smaller than neighborhood radius of c:
(5) tr is merged into c;
(6)else:
(7)if clusterNum>=MaxClusterNum:
(8) calculating the activity of each cluster;
(9) if there is no inactive cluster;
(10) merging the two clusters with the closest distance;
(11)clusterNum﹣=1;
(12)else:
(13) deleting the non-active cluster with the earliest average arrival time;
(14)clusterNum﹣=1;
(15) newly building a new cluster by taking tr as an element;
(16)clusterNum+=1;
the MaxClusternum is the number of the cluster types which can be accommodated by the local clustering module at most and is specified when the local clustering module is initialized. After the number of clusters in the local clustering module reaches MaxClusternum, a balanced state is achieved, and the centers of the clusters are stable under the condition that data change is not severe.
460. And sending a plurality of clustering centers obtained by the operation of each preset local clustering module to the global clustering module in real time.
470. And performing clustering operation on the received multiple clustering centers through the global clustering module to obtain a second set number of target clustering centers.
480. And evaluating the delivery effect of each delivered advertisement according to the second set number of target clustering centers and the feedback information.
According to the advertisement delivery effect evaluation method provided by the embodiment, the data compression processing is performed on the elements in each cluster type by adopting a characteristic vector mode, so that the data circulation and interaction speed is improved, and the data quality is ensured; if the new feedback information does not accord with the existing blending condition of various clusters blended into the current local clustering module, the number of clusters in the current local clustering module is kept unchanged by deleting the inactive clusters or combining the clusters and establishing a cluster by taking the new feedback information as the center, and the accuracy and the real-time performance of the line clustering module are ensured.
Example four
Fig. 5 is a schematic structural diagram of an apparatus for evaluating an advertisement delivery effect according to a fourth embodiment of the present invention. Referring to fig. 5, the apparatus comprises: an obtaining module 510, a distributing module 520, a first clustering module 530, a sending module 540, a second clustering module 550 and an evaluating module 560;
the obtaining module 510 is configured to obtain feedback information corresponding to each advertisement delivered in real time; the allocating module 520 is configured to allocate the feedback information to one of a plurality of preset local clustering modules based on a set rule; a first clustering module 530, configured to perform clustering operation on the feedback information through the local clustering module to obtain a first set number of clustering centers; a sending module 540, configured to send a plurality of clustering centers obtained through operation of preset local clustering modules to the global clustering module in real time; a second clustering module 550, configured to perform clustering operation on the received multiple clustering centers through the global clustering module to obtain a second set number of target clustering centers; the evaluation module 560 is configured to evaluate the delivery effect of each delivered advertisement according to the second set number of target cluster centers and the feedback information; the number of the preset local clustering modules is at least two, and the feedback information comprises attribute information of the delivered advertisements and demand information of the users for the delivered advertisements.
The device that advertisement putting effect aassessment that this embodiment provided, at first, carry out clustering operation to the feedback information of putting the advertisement respectively through distributed local clustering module, then cluster center that each distributed local clustering module operation obtained is clustered again by global clustering module, finally obtain target clustering center, the feedback information to each putting the advertisement according to target clustering center and putting the advertisement evaluates the effect of putting the advertisement, the purpose of carrying out real-time assessment to the effect of putting the advertisement has been realized, and then help advertisement operation personnel in time adjust the putting strategy, realize putting interests maximize, the mode through distributed clustering simultaneously, the problem of unit clustering inefficiency has been solved.
EXAMPLE five
Fig. 6 is a schematic structural diagram of an electronic device according to a third embodiment of the present invention. As shown in fig. 6, the electronic apparatus includes: a processor 670, memory 671, and computer programs stored on memory 671 and operable on processor 670; the number of the processors 670 may be one or more, and fig. 6 illustrates one processor 670 as an example; processor 670, when executing the computer program, implements the advertisement impression evaluation method as described in the above embodiments. As shown in fig. 6, the electronic device may further include an input device 672 and an output device 673. The processor 670, memory 671, input device 672 and output device 673 may be connected by a bus or other means, such as by a bus connection in fig. 6.
The memory 671 is used as a computer-readable storage medium for storing software programs, computer-executable programs, and modules, such as the means/modules for advertisement placement effectiveness evaluation (e.g., the obtaining module 510, the allocating module 520, etc. in the means for advertisement placement effectiveness evaluation) in the embodiments of the present invention. The processor 670 executes various functional applications and data processing of the electronic device by executing software programs, instructions, and modules stored in the memory 671, so as to implement the advertisement effectiveness evaluation method described above.
The memory 671 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, and an application program required for at least one function; the storage data area may store data created according to the use of the terminal, and the like. In addition, the memory 671 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid state storage device. In some examples, the memory 671 may further include memory located remotely from the processor 670, which may be connected to electronic devices/storage media through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The input device 672 may be used to receive input numeric or character information and generate key signal inputs related to user settings and function controls of the electronic apparatus. The output device 673 may include a display device such as a display screen.
EXAMPLE six
An embodiment of the present invention further provides a storage medium containing computer-executable instructions, which when executed by a computer processor, are configured to perform a method for evaluating advertisement delivery effectiveness, where the method includes:
acquiring feedback information corresponding to each advertisement in real time;
distributing the feedback information to one of a plurality of preset local clustering modules based on a set rule;
clustering operation is carried out on the feedback information through the local clustering module to obtain a first set number of clustering centers;
sending a plurality of clustering centers obtained by the operation of each preset local clustering module to a global clustering module in real time;
performing clustering operation on the received multiple clustering centers through the global clustering module to obtain a second set number of target clustering centers;
evaluating the delivery effect of each delivered advertisement according to the second set number of target clustering centers and the feedback information;
the number of the preset local clustering modules is at least two, and the feedback information comprises attribute information of the delivered advertisements and demand information of the users for the delivered advertisements.
Of course, the storage medium provided by the embodiment of the present invention contains computer-executable instructions, and the computer-executable instructions are not limited to the method operations described above, and may also perform operations related to the advertisement delivery effectiveness evaluation provided by any embodiment of the present invention.
From the above description of the embodiments, it is obvious for those skilled in the art that the present invention can be implemented by software and necessary general hardware, and certainly, can also be implemented by hardware, but the former is a better embodiment in many cases. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which may be stored in a computer-readable storage medium, such as a floppy disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a FLASH Memory (FLASH), a hard disk or an optical disk of a computer, and includes several instructions for enabling a computer device (which may be a personal computer, a storage medium, or a network device) to execute the embodiments of the present invention.
It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present invention and the technical principles employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.