Summary of the invention
The embodiment of the invention discloses a kind of information push method and device, low for the method source of exposure rate solving existing message push, and the shortcoming of waste platform resource; Concrete technical scheme is as follows:
A kind of information push method, be applied to message push platform, described method comprises:
PUSH message is treated in acquisition;
The interested first user group of PUSH message is treated in acquisition;
Obtain the behavioural characteristic of each user in described first user group;
According to the time of return disaggregated model of training in advance, in prediction first user group, each user's return messages pushes the time period divided in advance corresponding to the time interval of platform; Comprise in described time of return disaggregated model: the behavioural characteristic of user and user's return messages push the relation of the time period divided in advance corresponding to the time interval of platform;
Push the time period divided in advance corresponding to the time interval of platform according to each user's return messages in the first user group of prediction, obtain the time parameter values r that each user is correspondingi; Described time parameter values rithe length in the time interval of platform is pushed for identifying user return messages;
According to described time parameter riuser in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message;
Treat that PUSH message is pushed to determined potential user group by described.
In a kind of preferred implementation of this bright embodiment, the training process of described time of return disaggregated model comprises:
The time interval that double for user login message pushes platform is divided into T time period, T >=2;
Obtain M user the behavior record sample of specifying before the moment and after specifying the moment first return messages push the time interval of platform, M >=2;
Determine the user vector set that in M user, each user is corresponding, in described user vector set, comprise the time period divided in advance corresponding to the time interval of the identification information of this user, the behavioural characteristic extracted from this user behavior record sample and this user platform of return messages propelling movement first after specifying the moment;
Determined N number of user vector set is trained by the sorter preset, obtains time of return disaggregated model.
Wherein, described sorter comprises: the one in random forest, logistic regression and support vector machine classifier.
In a kind of preferred implementation of this bright embodiment, the corresponding interest parameter value d of each user in described first user groupi; Described interest parameter value difor identified user interest degree;
Described according to described time parameter riuser in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message, comprising:
According to described interest parameter value diwith described time parameter values riuser in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message.
Wherein, described according to described interest parameter value diwith described time parameter values riuser in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message, comprising:
According to interest parameter value diwith time parameter values ricalculate w1× di+ w2× ri, according to result of calculation, the user in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message; Wherein, w1for interest parameter value dicorresponding weight, w2for time parameter values ricorresponding weight.
Corresponding to embodiment of the method above, present invention also offers a kind of message push device, be applied to message push platform, described device comprises:
Message obtains module, treats PUSH message for obtaining;
First user group obtains module, treats the interested first user group of PUSH message for obtaining;
Behavioural characteristic obtains module, for obtaining the behavioural characteristic of each user in described first user group;
Time interval prediction module, for the time of return disaggregated model according to training in advance, in prediction first user group, each user's return messages pushes the time period divided in advance corresponding to the time interval of platform; Comprise in described time of return disaggregated model: the behavioural characteristic of user and user's return messages push the relation of the time period divided in advance corresponding to the time interval of platform;
Time parameter values obtains module, for pushing the time period divided in advance corresponding to the time interval of platform according to each user's return messages in the first user group of prediction, obtains the time parameter values r that each user is correspondingi; Described time parameter values rithe length in the time interval of platform is pushed for identifying user return messages;
Potential user group determination module, for according to described time parameter riuser in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message;
By described, message push module, for treating that PUSH message is pushed to determined potential user group.
In a kind of preferred implementation of this bright embodiment, also comprise the training module for training time of return disaggregated model, described training module comprises:
Time period divides submodule, is divided into T time period, T >=2 for the time interval double for user login message being pushed platform;
User behavior obtains submodule, for obtain M user the behavior record sample of specifying before the moment and after specifying the moment time interval of the platform of return messages propelling movement first, M >=2;
User vector set determination submodule, for the user vector set determining that in M user, each user is corresponding, in described user vector set, comprise the time period divided in advance corresponding to the time interval of the identification information of this user, the behavioural characteristic extracted from this user behavior record sample and this user platform of return messages propelling movement first after specifying the moment;
Model determination submodule, for determined N number of user vector set being trained by the sorter preset, obtains time of return disaggregated model.
Wherein, described sorter comprises: the one in random forest, logistic regression and support vector machine classifier.
In a kind of preferred implementation of this bright embodiment, the corresponding interest parameter value d of each user in described first user groupi; Described interest parameter value difor identified user interest degree;
Described potential user group determination module, specifically for:
According to described interest parameter value diwith described time parameter values riuser in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message.
Wherein, potential user group determination module, specifically for:
According to interest parameter value diwith time parameter values ricalculate w1× di+ w2× ri, according to result of calculation, the user in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message; Wherein, w1for interest parameter value dicorresponding weight, w2for time parameter values ricorresponding weight.
The technical scheme of the embodiment of the present invention, the possible return messages of the interest level and user for the treatment of PUSH message by user push the time interval two of platform because usually determining to treat the potential user group of PUSH message;
The potential user group determined by this method, not only to pushed message, there is higher interest-degree, there is again larger may logging in message valid time section simultaneously, this just improves the exposure rate of message effectively, reduce the situation that user does not log in storing message section effective time simultaneously, reduce the waste pushing platform resource.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, be clearly and completely described the technical scheme in the embodiment of the present invention, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, belong to the scope of protection of the invention.
The embodiment of the present invention provide firstly a kind of information push method, is applied to message push platform, and as shown in Figure 1, the method can comprise:
S101, obtains and treats PUSH message;
Saidly treat that PUSH message can be determined according to the service needed of message push platform, the specific implementation method of this step can adopt the related art scheme of prior art to realize, and the present invention does not do concrete restriction at this.
S102, obtains and treats the interested first user group of PUSH message;
The specific implementation method of this step can adopt the related art scheme of prior art to realize, and the present invention does not do concrete restriction at this.
Such as, in actual application, can determine that all users treat the interest level of PUSH message according to the rule preset; In prior art, for determining that the method for user to message interest level has a lot, the present invention does not do concrete restriction with this.
And then according to the threshold value preset choose most interested to message before S user be first user group.Said default threshold value can be pre-determined by research staff, and the present invention is in this no limit.
S103, obtains the behavioural characteristic of each user in described first user group;
In actual application, the behavioural characteristic of user can be obtained according to the behavior record of user; It should be noted that, the behavior record of said user is the behavior record relevant to message push platform.The behavioural characteristic obtained in this step is and the behavioural characteristic corresponding to time of return disaggregated model; Specifically, the behavioural characteristic obtained in this step should with training the behavioural characteristic utilized in time of return disaggregated model process to be corresponding.
Such as, when the behavioural characteristic utilized in training time of return disaggregated model process comprises: user at the appointed time in viewing amount of video, the channel of viewing; The behavioural characteristic so obtaining each user in described first user group also should comprise the at the appointed time interior viewing amount of video of user, the channel of viewing.
S104, according to the time of return disaggregated model of training in advance, in prediction first user group, each user's return messages pushes the time period divided in advance corresponding to the time interval of platform; Comprise in described time of return disaggregated model: the behavioural characteristic of user and user's return messages push the relation of the time period divided in advance corresponding to the time interval of platform;
In actual applications, the training process of said time of return disaggregated model can be as follows:
1, the time interval double for user login message being pushed platform is divided into T time period, T >=2;
The division of time period can according to treating that the actual needs of PUSH message is determined, if treat that PUSH message is more responsive to the time, time period division can be carried out in units of hour, day, if susceptibility is poor, also can by many days even time division sections in units of week, the moon etc.
Such as, when treating the ageing shorter namely more responsive to the time of PUSH message, the time period can be divided into: " within 5 hours ", " 5-12 hour ", " 12 hours-1 day ", " 1 day-2 days " and " more than 2 days ".
2, obtain M user the behavior record sample of specifying before the moment and after specifying the moment time interval of the platform of return messages propelling movement first, M >=2;
The quantity M of said user can be pre-determined by research staff.The present invention does not do concrete restriction with this.Be understandable that, M is larger, and the model obtained can be more accurate, but simultaneously, calculated amount also can increase.
The said appointment moment is also predetermined by research staff.When obtaining M user behavior record sample, M user can be obtained from the appointment moment, and the behavior record sample for message push platform before specifying the moment in certain period.
3, determine the user vector set that in M user, each user is corresponding, in described user vector set, comprise the time period divided in advance corresponding to the time interval of the identification information of this user, the behavioural characteristic extracted from this user behavior record sample and this user platform of return messages propelling movement first after specifying the moment;
In actual applications, class indication can be set for the time period divided in advance, such as, be divided into when the time period: when " within 5 hours ", " 5-12 hour ", " 12 hours-1 day ", " 1 day-2 days " and " more than 2 days ", " within 5 hours " can be arranged and be designated 1, " 5-12 hour " arranges and is designated 2, and by that analogy, last " more than 2 days " arrange and are designated 5.
In this case, M user vector set can be expressed as:
{ user id1, feature 1, feature 2, feature 3....1};
{ user id2, feature 1, feature 2, feature 3....4};
....
{ user idm, feature 1, feature 2, feature 3....5};
In above-mentioned user vector set, the first unique identification information being classified as user, last is classified as the class indication of the time period divided in advance corresponding to the time interval of user;
It should be noted that, in user vector set, the order of each element is can be predetermined by research staff, and the identification information of user can at secondary series, or in other row.But element corresponding in each user vector set should be in the same position in each user vector set.Such as, if the unique identification information of user is arranged in the second of user vector set, so the unique identification information of M user all should be arranged in the second of each user vector set.
The behavioural characteristic extracted from user behavior record sample, can be determined by research staff, the present invention does not do concrete restriction at this in advance according to actual needs.Be understandable that, the behavioural characteristic corresponding to different types of message push platform may be different.
4, determined N number of user vector set is trained by the sorter preset, obtain time of return disaggregated model.
Said sorter can adopt correlation classifier of the prior art, such as, and random forest sorter, logistic regression sorter or support vector machine classifier.Concrete form the present invention of sorter is in this no limit, can be selected according to actual needs by research staff.
In actual applications, can also determine that the method for user vector set gathers other user vector set multiple again with above-mentioned, for carrying out the test of generalization ability to the time of return disaggregated model obtained.
After obtaining time of return disaggregated model, according to the behavioural characteristic of each user obtained in step S103, just can predict that each user's return messages pushes the time period divided in advance corresponding to the time interval of platform.
S105, pushes the time period divided in advance corresponding to the time interval of platform according to each user's return messages in the first user group of prediction, obtains the time parameter values r that each user is correspondingi; Described time parameter values rithe length in the time interval of platform is pushed for identifying user return messages;
In actual applications, can pre-determine divided time period and the corresponding relation of score value, and the time that the return messages that represent of the time period divided push platform is shorter, its score value is higher; Such as, be divided into when the time period: when " within 5 hours ", " 5-12 hour ", " 12 hours-1 day ", " 1 day-2 days " and " more than 2 days ", " within 5 hours " represent that this user probably again returned platform in following 5 hours, this time period will a corresponding higher score value, the score value that " 5-12 hour " is corresponding is lower slightly, by that analogy, the score value that " more than 2 days " are corresponding is minimum.
The time parameter values r that user is correspondingithese user's return messages predicted by time of return disaggregated model exactly push the score value of the time period divided in advance corresponding to the time interval of platform.
S106, according to described time parameter riuser in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message;
In actual applications, the embodiment of this step can be: by the time parameter r of each user in first user groupiarranging from small to large, then pushing customer volume, from time parameter r according to presettingiminimum user starts, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message; Such as, presetting and pushing customer volume is 10, then select time parameter ri10 minimum users are as the described potential user group treating PUSH message.
In addition, in actual applications, in first user group, each user can a corresponding interest parameter value di; Described interest parameter value difor identified user interest degree;
Now, step S106 can also be: according to described interest parameter value diwith described time parameter values riuser in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message.
Concrete, can according to interest parameter value diwith time parameter values ricalculate w1× di+ w2× ri, according to result of calculation, the user in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message; Wherein, w1for interest parameter value dicorresponding weight, w2for time parameter values ricorresponding weight.
At employing w1× di+ w2× riresult when sorting, also result can being arranged from small to large, then pushing customer volume according to presetting, from the user that result is minimum, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message; Such as, presetting and pushing customer volume is 10, then 10 users that selection result is minimum are as the described potential user group treating PUSH message.
It should be noted that, w1and w2can be determined according to service needed by research staff, the present invention does not do concrete restriction at this.
It should be noted that further, for interest parameter value di, the level of interest that user treats PUSH message is higher, interest parameter value dilarger.Interest parameter value didefining method, can adopt the related art scheme of prior art to realize, the present invention does not do concrete restriction at this.Such as, while the interested first user group of PUSH message is treated in step S102 acquisition, the interest parameter value d of each user in first user group can just be determinedi.
By described, S107, treats that PUSH message is pushed to determined potential user group.
This step can adopt the related art scheme of prior art to realize, and the present invention does not do concrete restriction at this.
Can be found out by above-mentioned embodiment of the method, the technical scheme of the embodiment of the present invention, the possible return messages of the interest level and user for the treatment of PUSH message by user push the time interval two of platform because usually determining to treat the potential user group of PUSH message;
The potential user group determined by this method, not only to pushed message, there is higher interest-degree, there is again larger may logging in message valid time section simultaneously, this just improves the exposure rate of message effectively, reduce the situation that user does not log in storing message section effective time simultaneously, reduce the waste pushing platform resource.
Corresponding to embodiment of the method above, present invention also offers a kind of message push device, as shown in Figure 2, be applied to message push platform, described device comprises:
Message obtains module 101, treats PUSH message for obtaining;
First user group obtains module 102, treats the interested first user group of PUSH message for obtaining;
Behavioural characteristic obtains module 103, for obtaining the behavioural characteristic of each user in described first user group;
Time interval prediction module 104, for the time of return disaggregated model according to training in advance, in prediction first user group, each user's return messages pushes the time period divided in advance corresponding to the time interval of platform; Comprise in described time of return disaggregated model: the behavioural characteristic of user and user's return messages push the relation of the time period divided in advance corresponding to the time interval of platform;
Time parameter values obtains module 105, for pushing the time period divided in advance corresponding to the time interval of platform according to each user's return messages in the first user group of prediction, obtains the time parameter values ri that each user is corresponding; Described time parameter values ri is used for the length that identifying user return messages push the time interval of platform;
Potential user group determination module 106, for sorting to the user in first user group according to described time parameter ri, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message;
By described, message push module 107, for treating that PUSH message is pushed to determined potential user group.
In actual applications, this device can also comprise the training module for training time of return disaggregated model, and described training module comprises:
Time period divides submodule, is divided into T time period, T >=2 for the time interval double for user login message being pushed platform;
User behavior obtains submodule, for obtain M user the behavior sample of specifying before the moment and after specifying the moment time interval of the platform of return messages propelling movement first, M >=2;
User vector set determination submodule, for the user vector set determining that in M user, each user is corresponding, in described user vector set, comprise the time period divided in advance corresponding to the time interval of the identification information of this user, the behavioural characteristic extracted from this user behavior sample and this user platform of return messages propelling movement first after specifying the moment;
Model determination submodule, for determined N number of user vector set being trained by the sorter preset, obtains time of return disaggregated model.
Wherein, described sorter can comprise: the one in random forest, logistic regression and support vector machine classifier.
In the embodiment of above-mentioned each device, the corresponding interest parameter value di of each user in described first user group; Described interest parameter value di is used for identified user interest degree;
Described potential user group determination module 106, specifically may be used for:
According to described interest parameter value di and described time parameter values ri, the user in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message.
In actual applications, w1 × di+w2 × ri can be calculated according to interest parameter value di and time parameter values ri, according to result of calculation, the user in first user group is sorted, according to ranking results, select with preset push customer volume respective numbers user as the described potential user group treating PUSH message; Wherein, w1 is the weight that interest parameter value di is corresponding, and w2 is the weight that time parameter values ri is corresponding.
Can be found out by above-mentioned embodiment of the method, the technical scheme of the embodiment of the present invention, the possible return messages of the interest level and user for the treatment of PUSH message by user push the time interval two of platform because usually determining to treat the potential user group of PUSH message;
The potential user group determined by this method, not only to pushed message, there is higher interest-degree, there is again larger may logging in message valid time section simultaneously, this just improves the exposure rate of message effectively, reduce the situation that user does not log in storing message section effective time simultaneously, reduce the waste pushing platform resource.
It should be noted that, for device embodiment, because it is substantially similar to embodiment of the method, so description is fairly simple, relevant part illustrates see the part of embodiment of the method.
For convenience of description, various module is divided into describe respectively with function when describing above device.Certainly, the function of each module can be realized in same or multiple software and/or hardware when implementing of the present invention.
It should be noted that, in this article, the such as relational terms of first and second grades and so on is only used for an entity or operation to separate with another entity or operational zone, and not necessarily requires or imply the relation that there is any this reality between these entities or operation or sequentially.And, term " comprises ", " comprising " or its any other variant are intended to contain comprising of nonexcludability, thus make to comprise the process of a series of key element, method, article or equipment and not only comprise those key elements, but also comprise other key elements clearly do not listed, or also comprise by the intrinsic key element of this process, method, article or equipment.When not more restrictions, the key element limited by statement " comprising ... ", and be not precluded within process, method, article or the equipment comprising described key element and also there is other identical element.
Each embodiment in this instructions all adopts relevant mode to describe, between each embodiment identical similar part mutually see, what each embodiment stressed is the difference with other embodiments.Especially, for device embodiment, because it is substantially similar to embodiment of the method, so description is fairly simple, relevant part illustrates see the part of embodiment of the method.
One of ordinary skill in the art will appreciate that all or part of step realized in said method embodiment is that the hardware that can carry out instruction relevant by program has come, described program can be stored in computer read/write memory medium, here the alleged storage medium obtained, as: ROM/RAM, magnetic disc, CD etc.
The foregoing is only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.All any amendments done within the spirit and principles in the present invention, equivalent replacement, improvement etc., be all included in protection scope of the present invention.