CN110727859B - Recommendation information pushing method and device - Google Patents
Recommendation information pushing method and device Download PDFInfo
- Publication number
- CN110727859B CN110727859B CN201910867360.4A CN201910867360A CN110727859B CN 110727859 B CN110727859 B CN 110727859B CN 201910867360 A CN201910867360 A CN 201910867360A CN 110727859 B CN110727859 B CN 110727859B
- Authority
- CN
- China
- Prior art keywords
- user
- recommendation
- recommendation information
- information
- label
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Abstract
The invention provides a recommendation information pushing method and a recommendation information pushing device, and belongs to the technical field of computer information. The recommendation information pushing method provided by the invention comprises a data establishing stage and a topic distributing stage, wherein in the data establishing stage, users are classified into a plurality of user groups according to operation records of the users, and personal recommendation labels of the users and group recommendation labels of the user groups are screened out. In the information pushing stage, matched recommendation information is screened out according to the personal recommendation label and the group recommendation label of the user, and then the recommendation information is further screened out according to the position and time of the user for pushing, so that recommendation information which the user is interested in can be more accurately selected for pushing, and the browsing experience of the user is improved.
Description
Technical Field
The invention belongs to the technical field of computer information, and particularly relates to a recommendation information pushing method and a recommendation information pushing device.
Background
Network shopping and network communities are more and more widely entering the lives of people, and when a user browses information of the network communities or browses commodities of shopping websites, the websites can often automatically push some information or commodities to the user. However, in the prior art, information or goods can only be pushed based on personal browsing data of a user, and because the reference factor is too single, the interest degree of the user in the information or goods pushed by the website is low, thereby affecting the browsing experience of the user.
Disclosure of Invention
The invention aims to at least solve one of the technical problems in the prior art, and provides a recommendation information pushing method, which can select recommendation information interested by a user more accurately to push by screening the recommendation information by combining multiple factors, so that the browsing experience of the user is improved.
The technical scheme adopted for solving the technical problem of the invention is a recommendation information pushing method, which comprises the following steps:
a data establishing stage:
establishing a recommendation information database, wherein each recommendation information comprises at least one label;
acquiring operation records of each user, and respectively generating associated data of each user according to a preset first algorithm according to the operation records of each user; the associated data is the relation between the label included in each piece of recommendation information operated by the user and the operation record;
classifying the users into a plurality of user groups according to the associated data of the users and a classifier obtained by pre-training;
according to the associated data of each user, respectively screening out the group recommendation labels of each user group and the individual recommendation labels of each user in each user group according to a preset second algorithm;
and an information pushing stage:
acquiring the user group to which the user belongs, and screening the recommendation information matched with the label in the recommendation information database according to the group recommendation label of the user group to which the user belongs and the personal recommendation label of the user;
obtaining position information of a user, determining time for pushing the recommendation information, calculating recommendation scores of the selected recommendation information according to a preset third algorithm by combining the position information and the time, arranging the recommendation information according to a sequence of the recommendation scores from high to low, and selecting the first K pieces of recommendation information to push to the user.
According to the method provided by the invention, the recommendation information is screened out according to the personal operation records of the user and the operation records of each user in the user group to which the user belongs, and then the recommendation information is further screened out by combining the factors such as the geographical position, the time and the like of the user and pushed to the user, so that the recommendation information which the user is interested in can be more accurately selected and pushed, and the browsing experience of the user is improved.
Preferably, in the method provided by the present invention, the recommendation information includes a commodity and a content; the operation records comprise purchase records of the commodities, praise records, comment records and forwarding records of the contents; for each tag in the associated data, the preset first algorithm satisfies:
A1×lg(B1)+A2×lg(B2)+A3×lg(B3)+A4×lg(C1)
wherein B1 contains the number of tags for each user's approved content; a1 is a weight for agreeing on the content;
b2 contains the number of tags for each user-reviewed content; a2 is the weight for commenting on content;
b3 contains the number of labels for each user's forwarded content; a3 is the weight for forwarding the content;
c1 contains the number of labels for each item purchased by the user; a4 is the weight for purchasing the commodity.
Preferably, in the above method provided by the present invention, the pre-trained classifier uses a clustering algorithm to classify each user into a plurality of user groups.
Preferably, in the method provided by the present invention, the screening out the group recommendation labels of each user group and the individual recommendation labels of each user in each user group according to a preset second algorithm specifically includes:
screening a group recommendation label:
acquiring the number of labels contained in the recommendation information in the association data of each user in each user group, arranging the labels in a sequence from high to low, and selecting the first J labels as group recommendation labels of the user group;
screening personal recommendation labels:
and acquiring the number of each label contained in the recommendation information in the personal association data of the user, arranging the labels according to the sequence of the number from high to low, and selecting the first L labels as the personal recommendation labels of the user.
Preferably, in the above method provided by the present invention, the location information includes a plurality of tags set according to location scenarios, the time includes a plurality of tags set according to time scenarios, and each location information and each tag at each time have a corresponding weight; each recommendation information has the same basic recommendation score; the calculating, by combining the position information and the time, a recommendation score of the screened recommendation information according to a preset third algorithm specifically includes:
obtaining the screened labels of each piece of recommendation information, comparing whether the label of each piece of recommendation information is the same as the label under the position information of the user or not, and comparing whether the label of each piece of recommendation information is the same as the label under the time of the user or not;
if the label exists, multiplying the weight of the same label with the basic recommendation score of the recommendation information respectively, and taking the sum of the products as the basic recommendation score of the recommendation information.
Correspondingly, the invention also provides a recommendation information pushing device, which comprises: the device comprises a data establishing unit and an information pushing unit;
the data establishing unit specifically includes:
the database building module is used for building a recommendation information database, and each recommendation information comprises at least one label;
the associated data generation module is used for acquiring operation records of each user and respectively generating associated data of each user according to a preset first algorithm according to the operation records of each user; the associated data is the relation between the label included in each piece of recommendation information operated by the user and the operation record;
the user classification module is used for classifying the users into a plurality of user groups according to the associated data of the users and a classifier obtained by pre-training;
the recommendation label calculation module is used for respectively screening out the group recommendation labels of all user groups and the individual recommendation labels of all users in each user group according to the associated data of all users and a preset second algorithm;
the information pushing unit specifically includes:
the recommendation information screening module is used for acquiring the user group to which the user belongs, and screening recommendation information matched with the labels in the recommendation information database according to the group recommendation label of the user group to which the user belongs and the personal recommendation label of the user;
and the recommendation information scoring module is used for acquiring the position information of the user, determining the time for pushing the recommendation information, calculating the recommendation scores of the selected recommendation information according to a preset third algorithm by combining the position information and the time, arranging the recommendation information according to a sequence of the recommendation scores from high to low, and selecting the first K pieces of recommendation information to push to the user.
Preferably, in the above apparatus provided by the present invention, the recommendation information includes a commodity and a content; the operation records comprise purchase records of the commodities, praise records, comment records and forwarding records of the contents; in the associated data generation module, for each tag in the associated data, the preset first algorithm satisfies:
A1×lg(B1)+A2×lg(B2)+A3×lg(B3)+A4×lg(C1)
wherein B1 contains the number of tags for each user's approved content; a1 is a weight to like content;
b2 contains the number of tags for each user-reviewed content; a2 is the weight for commenting on content;
b3 contains the number of labels for each user's forwarded content; a3 is the weight for forwarding the content;
c1 contains the number of labels for each item purchased by the user; a4 is the weight for purchasing the commodity.
Preferably, in the apparatus according to the present invention, in the user classification module, the pre-trained classifier classifies each user into a plurality of user groups by using a clustering algorithm.
Preferably, in the above apparatus provided by the present invention, the recommended label calculation module specifically includes: the system comprises a group recommendation tag module and a personal recommendation tag module;
the group recommendation tags are used for acquiring the number of the tags contained in the recommendation information in the association data of each user in each user group, arranging the tags in a high-to-low order, and selecting the first J tags as the group recommendation tags of the user group;
the personal recommendation tags are used for acquiring the number of each tag contained in the recommendation information in the personal association data of the user, arranging the tags in a sequence from high to low, and selecting the first L tags as the personal recommendation tags of the user.
Preferably, in the above apparatus provided by the present invention, the position information includes a plurality of tags set according to position scenes, the time information includes a plurality of tags set according to time scenes, and each position information and each tag at time have a corresponding weight; each recommendation information has the same basic recommendation score;
the recommendation information scoring module is specifically configured to obtain each of the screened labels of the recommendation information, compare whether the label of each of the recommendation information is the same as the label of the user in the location information, and compare whether the label of each of the recommendation information is the same as the label of the user in the time;
if the label exists, multiplying the weight of the same label by the basic recommendation score of the recommendation information respectively, and taking the sum of the products as the basic recommendation score of the recommendation information.
Drawings
Fig. 1 is a flowchart of a data establishment phase in a recommendation information pushing method provided in this embodiment;
fig. 2 is a flowchart of an information push phase in the method for pushing recommended information according to this embodiment;
fig. 3 is a schematic structural diagram of a recommendation information pushing device provided in this embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings. 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.
The shapes and sizes of the various elements in the drawings are not to scale and are merely intended to facilitate an understanding of the contents of the embodiments of the invention.
The embodiment provides a recommendation information pushing method, which comprises the following steps:
as shown in fig. 1, during the data set-up phase:
and S11, establishing a recommendation information database, wherein each recommendation information comprises at least one label.
Specifically, a recommendation information database is established as required, the recommendation information may include multiple types, each type has multiple pieces of recommendation information of the type, and at least one tag is added to each piece of recommendation information according to the specific content of each piece of recommendation information. For example, the recommendation information includes two types of commodities and contents, and if a green-plant commodity exists under the commodity type, labels such as 'plants', 'green', 'home decoration', and the like can be established for the commodity. The specific categories and labels may be designed as desired and are not limited herein.
And S12, acquiring the operation records of each user, and respectively generating the associated data of each user according to the operation records of each user and a preset first algorithm.
The associated data is a relationship between a tag included in each piece of recommendation information operated by the user and the operation record.
Specifically, each tag included in the recommendation information operated by the user is acquired, the recommendation information may include multiple types, and the number of times that the user has operated each type of recommendation information under each tag can be obtained according to the operation record of the user. And the user can respectively perform several different operations on each type of recommendation information, so that the times of each operation performed on each type of recommendation information by the user under each label can be obtained according to the operation records of the user, a weight is respectively assigned to each operation, and the preset first algorithm is that the relation between each label and the operation record of the user, namely the associated data of the user is obtained according to the weight of the operation.
For example, the recommendation information includes two types of goods and contents, the operation type of the goods by the user includes purchase, the operation type of the contents by the user includes approval, comment and forwarding, the operation record includes a purchase record of the goods, and approval, comment and forwarding records of the contents. If the operated goods or contents include N tags in the operation record of the user, the preset first algorithm satisfies, for each tag in the associated data generated according to the operation record of the user:
A1×lg(B1)+A2×lg(B2)+A3×lg(B3)+A4×lg(C1)
wherein B1 contains the number of tags for each user's approved content; a1 is the weight that favors the content.
B2 contains the number of tags for each user-reviewed content; a2 is the weight for commenting on content.
B3 contains the number of labels for each user's forwarded content; a3 is the weight for forwarding the content.
C1 contains the number of labels for each item purchased by the user; a4 is the weight for purchasing the commodity.
If the user has performed only one type of recommendation information or only one type of recommendation information under one tag, the right of the operation may be calculated only in the above equation. The right item of the recommendation information that has not been operated is 0.
And by analogy, the N labels included in the commodities or contents operated by the user are calculated according to the formula, and the results of the N labels are synthesized into the associated data table of the user. For example, the commodity or content operated by the user a includes a tag P and a tag Q, and the user approves the content including the tag P4 times, reviews the content 3 times, and forwards the content 1 time; the user has purchased the commodity containing the tag P2 times; the user approves the content containing the label Q for 7 times, reviews the content for 2 times and forwards the content for 0 time; the user has purchased the item containing tag Q5 times. The associated data table for the user is as follows:
| user' s | Label (R) | Relation formula |
| P | M | A1×lg(4)+A2×lg(3)+A3×lg(1)+A4×lg(2) |
| Q | N | A1×lg(7)+A2×lg(2)+A4×lg(5) |
Of course, the types included in the recommendation information and the operation that can be performed on each type of recommendation information may be in other forms, for example, the above example may also record a comment record for commenting on purchased commodities, and the specific setting may be performed as needed, which is not limited herein. The algorithm of the associated data may be other algorithms, and is specifically designed according to the needs, and is not limited herein.
Optionally, the associated data of the user may be updated in real time according to the change of the operation record of the user.
And S13, classifying the users into a plurality of user groups according to the associated data of the users and the classifier obtained by pre-training.
Specifically, according to the associated data of each user, the classifier obtained by pre-training may classify each user into a plurality of user groups by using a clustering algorithm. The classifier acquires the associated data of each user, takes the associated data of each user as input data, takes each label as a dimension, selects a proper metric, such as Euclidean metric or Manhattan distance metric, and performs cluster analysis on each user to obtain a plurality of data clusters, wherein each data cluster is each user group, and each data in the data clusters is each user. According to the operation of the user on the recommendation information under each label, the user is divided into a plurality of user groups, namely in the same user group, the operation similarity of each user on the recommendation information is highest, so that the user can be classified more accurately, and the condition that the classification of the user is inaccurate due to the fact that a single variable is used for classifying the user is avoided. Of course, other methods may also be used to classify the users, and the specific method may be designed according to actual needs, which is not limited herein.
Optionally, the classifier obtained by pre-training may adopt various types of clustering algorithms to classify the users, for example, any one of a K-means clustering algorithm, a center point clustering algorithm, and a random selection clustering algorithm. Taking the example of classifying users by adopting a K-means clustering algorithm, if the users need to be divided into K user groups, establishing a multi-dimensional coordinate system by using the associated data of each user, firstly randomly selecting the associated data of K users as an initial clustering center, then calculating the distance between the associated data of each user and the associated data of each user serving as the clustering center, and allocating each user to the user closest to the user serving as the clustering center according to the calculation. And if all the users are distributed, K data clusters are obtained, and then new K users serving as clustering centers are calculated according to the position of each user in the K data clusters. This process is repeated until a termination condition is met, which can be set as desired, e.g., no (or a minimum number) of users are reassigned to different data clusters, or no (or a minimum number) of cluster centers of the data clusters change. Specifically, the design is as required, and is not limited herein.
Further, the user group of each user may be updated according to the change of the associated data of the user.
And S14, respectively screening the group recommendation labels of the user groups and the individual recommendation labels of the users in each user group according to the associated data of the users and a preset second algorithm.
Specifically, the screening of the group recommendation label in S14 may include:
the number of the labels contained in the recommendation information in the association data of each user in each user group is obtained, the labels are arranged according to the sequence from high to low, the more the labels are, the more the user in the user group is interested in the recommendation information comprising the labels, the first J labels can be selected as the group recommendation labels of the user group. For example, the user group H includes 300 users, obtains the association data of the 300 users, counts up the tags associated in the recommendation information included in the association data of the 300 users, and the number of each tag, for example, the associated data of 300 users includes 5 tags, which are marked as tags one to five, and the associated data of 300 users includes recommended information tags, the number of the first label is 100 times, the number of the second label is 30 times, the number of the third label is 50 times, the number of the fourth label is 70 times, the number of the fifth label is 20 times, then tag one to tag five are arranged in order of the number from high to low [ tag one, tag four, tag three, tag two, tag five ], J may be 2, namely, the first 2 tags, i.e. tag one and tag four, are selected as the group recommendation tags of the user group H.
Further, the screening of the personal recommendation label in S14 may include:
the number of each label contained in the recommendation information in the personal association data of the user is obtained, the labels are arranged according to the sequence from high to low, the more labels, the more interest of the user in the recommendation information including the label, the first L labels can be selected as the personal recommendation label of the user.
For example, obtaining the associated data of the user a, the user a operates 10 pieces of recommendation information, and counts each tag included in the 10 pieces of recommendation information, and the number of each tag, where for example, the 10 pieces of recommendation information include 3 tags, which are denoted as tag S, tag Z, and tag X, and in the recommendation information operated by the user a, the number of the tag S is 5 times, the number of the tag Z is 3 times, and the number of the tag X is 6 times, then the tags S, tag Z, and tag X are arranged in order from high to low in number as [ tag X, tag S, tag Z ], and L may be 1, that is, the first 1 tag is selected, that is, the tag X is selected as the personal recommendation tag of the user a.
Optionally, the group recommendation tag and the individual recommendation tag may be updated in real time according to real-time operation of each user in each group on recommendation information.
Optionally, if the number of users is large and the generated operation data is large, stream computing (stream computing) may be used to update the group recommendation tag in real time, so as to quickly respond to changes in user operations. Specifically, for example, a streaming framework such as Storm or spark streaming can be used for the calculation. Of course, other types of streaming frameworks may be used for calculation, and the calculation may be specifically designed according to actual needs, and is not limited herein.
As shown in fig. 2, in the information push phase:
s21, obtaining the user group to which the user belongs, and screening recommendation information matched with the label in the recommendation information database according to the group recommendation label of the user group to which the user belongs and the personal recommendation label of the user.
Specifically, the recommendation information database comprises a plurality of recommendation information, each recommendation information comprises at least one tag, and according to the personal recommendation tag of the user and the tags contained in the group recommendation tags to which the user belongs, recommendation information of the same tag can be screened out from the recommendation information database to serve as recommendation information which is possibly interested by the user.
Optionally, different weights may be set for the group recommendation tag and the individual recommendation tag, for example, the weight of the individual recommendation tag is greater than the weight of the group recommendation tag, recommendation information screened in the recommendation information database by the individual recommendation tag is processed differently according to the weights, and recommendation information screened in the recommendation information database by the group recommendation tag is processed differently, so as to obtain more accurate recommendation information in which the user is interested.
The recommendation information is screened according to the group recommendation tags of the user group to which the user belongs, the recommendation information which is possibly interested by the user can be predicted in advance, the recommendation information is screened according to the personal recommendation tags of the user, the user and the interested recommendation information can be known, the recommendation information which is interested by the user group to which the user belongs is integrated, and the recommendation information which is interested by the user is screened according to the recommendation information which is interested by the user, so that the recommendation information which is interested by the user can be selected more accurately to be pushed, and the browsing experience of the user is improved.
S22, obtaining the position information of the user, determining the time for pushing the recommendation information, calculating the recommendation score of the screened recommendation information according to a preset third algorithm by combining the position information and the time, arranging the recommendation information according to the sequence of the recommendation scores from high to low, and pushing the first K pieces of recommendation information to the user.
Specifically, a location information group including a plurality of location information items may be established, each location information item includes a plurality of tags set according to a location scene, a time point group including a plurality of times may be established, each time item includes a plurality of tags set according to a time scene, the tags in each location information item are respectively assigned with a corresponding weight, and the tags in each time item are respectively assigned with a corresponding weight. The same basic recommendation score is set for each recommendation information. In S22, the recommendation score of the recommendation information screened in S21 is calculated according to a preset third algorithm in combination with the position information and the time, and specifically includes:
obtaining the tag of each piece of the recommended information screened in the step S21, comparing whether the tag of each piece of the recommended information is the same as the tag of the position information of the user, and comparing whether the tag of each piece of the recommended information is the same as the tag of the user in time.
If the label exists, the obtained weight of the same label is multiplied by the basic recommendation score of the recommendation information, and the sum of the products is used as the basic recommendation score of the recommendation information.
For example, the location information group includes a subway, an office, and a mall; the time point group comprises 8-12 am, 12-19 pm and 19-24 pm. The subway comprises related labels of entertainment and music, and the office comprises labels of work and storage tools. The morning includes related labels of 'breakfast', 'sports', and the evening includes labels of 'dinner', 'sleep aid', and the like. The recommendation information of the user a screened in S21 may include, for example, coffee (goods) and sports music recommendations (contents), where the label included in coffee is "drink" or "work", the label included in the sports music recommendations is "sports", the current location information of the user a is obtained as an office, the time when the recommendation information is pushed to the user a this time is determined to be morning, and then the time when the recommendation information is pushed to the user a (coffee and sports music recommendations) is obtained by comparison, where the recommendation score of the coffee is the basic recommendation score of the coffee, plus the product of the basic recommendation score of the coffee and the weight of the label "work", and the location information (office) of the user has the same label "work" as the location information (office) of the user, and the time when the recommendation information is pushed to the user a morning is determined to be morning; the recommendation score for a sports music recommendation is the base recommendation score for the sports music recommendation, plus the product of its base recommendation branch and the weight of the tag "sports". And comparing the obtained recommended value of the coffee with the recommended value recommended by the sports music, and taking a piece of higher recommended information as final recommended information to be pushed to the user.
Of course, the specific location information and the time may be designed according to the need, and are not limited herein.
Optionally, if the recommendation information pushing method provided in this embodiment is applied to a website or an application program, when a user logs in the website or the application program for the first time, the initial association tag of the user may be obtained according to the login information of the user, such as age and gender, and the content in the bound social account, and a page including multiple tags may also be pushed to the user, so that the user selects a tag that is of interest to the user as the initial association tag of the user.
In summary, according to the recommendation information pushing method provided by this embodiment, because the recommendation information is screened out according to the operation record of the user and the operation record of each user in the user group to which the user belongs, and then the recommendation information is further screened out by combining the geographic position, the time and other factors of the user and pushed to the user, the recommendation information that the user is interested in can be more accurately selected for pushing, and the browsing experience of the user is improved.
Correspondingly, as shown in fig. 3, the present embodiment further provides a recommendation information pushing device, which includes: a data establishing unit 1 and an information pushing unit 2,
specifically, the data establishment unit 1 may include:
and a database building module 11, configured to build a recommendation information database, where each recommendation information includes at least one tag.
And the associated data generating module 12 is configured to obtain operation records of each user, and generate associated data of each user according to a preset first algorithm according to the operation records of each user. The associated data is the relationship between the label included in each piece of recommended information operated by the user and the operation record.
And a user classification module 13, configured to classify each user into a plurality of user groups according to the associated data of each user and a classifier obtained through pre-training.
And the recommended label calculation module 14 is configured to respectively screen out the group recommended labels of each user group and the individual recommended labels of each user in each user group according to a preset second algorithm according to the associated data of each user.
Specifically, the information pushing unit 2 may include:
and the recommendation information screening module 21 is configured to acquire a user group to which the user belongs, and screen recommendation information matched with the tag in the recommendation information database according to the group recommendation tag of the user group to which the user belongs and the personal recommendation tag of the user.
And the recommendation information scoring module 22 is configured to obtain the position information of the user, determine time for pushing the recommendation information, calculate recommendation scores of the screened recommendation information according to a preset third algorithm by combining the position information and the time, rank the recommendation information according to a sequence of the recommendation scores from high to low, and select the first K pieces of recommendation information to push to the user.
Optionally, in the apparatus provided in this embodiment, the recommendation information includes a product and a content. The operation records comprise purchase records of commodities, approval records, comment records and forwarding records of contents. In the associated data generating module 12, for each tag in the associated data, a preset first algorithm satisfies:
A1×lg(B1)+A2×lg(B2)+A3×lg(B3)+A4×lg(C1)
wherein B1 contains the number of tags for each user's approved content; a1 is a weight for agreeing on the content;
b2 contains the number of tags for each user-reviewed content; a2 is the weight for commenting on content;
b3 contains the number of labels for each user's forwarded content; a3 is the weight for forwarding the content;
c1 contains the number of labels for each item purchased by the user; a4 is the weight for purchasing the commodity.
Optionally, in the apparatus provided in this embodiment, in the user classification module 13, a classifier trained in advance uses a clustering algorithm to classify each user into a plurality of user groups.
Optionally, in the apparatus provided in this embodiment, the recommended label calculating module 14 may specifically include: a group recommendation tag module 141 and a personal recommendation tag module 142.
The group recommendation tag 141 is configured to obtain the number of each tag included in the recommendation information in the association data of each user in each user group, arrange the tags in a descending order of the number, and select the first J tags as the group recommendation tags of the user group.
The personal recommendation tags 142 are used for acquiring the number of each tag included in the recommendation information in the personal association data of the user, arranging the tags in a high-to-low order, and selecting the first L tags as the personal recommendation tags of the user.
Optionally, in the above apparatus provided in this embodiment, the position information includes a plurality of tags set according to the position scene, the time information includes a plurality of tags set according to the time scene, and each of the position information and the tags at each time have a corresponding weight. Each recommendation information has the same base recommendation score. The recommendation information scoring module 22 is specifically configured to obtain a tag of each piece of screened recommendation information, compare whether the tag of each piece of recommendation information and the tag under the location information of the user have the same tag, compare whether the tag of each piece of recommendation information and the tag under the time of the user have the same tag, if yes, multiply the weight of the same tag by the basic recommendation score of the piece of recommendation information, and use the sum of the products as the basic recommendation score of the piece of recommendation information.
It will be understood that the above embodiments are merely exemplary embodiments taken to illustrate the principles of the present invention, which is not limited thereto. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit and substance of the invention, and these modifications and improvements are also considered to be within the scope of the invention.
Claims (6)
1. A recommendation information pushing method is characterized by comprising the following steps:
a data establishing stage:
establishing a recommendation information database, wherein each recommendation information comprises at least one label;
acquiring operation records of each user, and respectively generating associated data of each user according to a preset first algorithm according to the operation records of each user; the associated data is the relation between the label included in each piece of recommendation information operated by the user and the operation record;
classifying the users into a plurality of user groups according to the associated data of the users and a classifier obtained by pre-training;
respectively screening out the group recommendation labels of each user group and the individual recommendation labels of each user in each user group according to the associated data of each user and a preset second algorithm;
and an information pushing stage:
acquiring the user group to which the user belongs, and screening recommendation information matched with the label in the recommendation information database according to the group recommendation label of the user group to which the user belongs and the personal recommendation label of the user;
acquiring position information of a user, determining time for pushing the recommendation information, calculating recommendation scores of the screened recommendation information according to a preset third algorithm by combining the position information and the time, arranging the recommendation information according to a sequence of the recommendation scores from high to low, and selecting the first K pieces of recommendation information to push to the user;
the recommendation information comprises commodities and contents; the operation records comprise purchase records of the commodities, praise records, comment records and forwarding records of the contents; for each tag in the associated data, the preset first algorithm satisfies:
A1×lg(B1)+A2×lg(B2)+A3×lg(B3)+A4×lg(C1)
wherein B1 contains the number of tags for each user's approved content; a1 is a weight for agreeing on the content;
b2 contains the number of tags for each user-reviewed content; a2 is the weight for commenting on content;
b3 contains the number of labels for each user's forwarded content; a3 is the weight for forwarding the content;
c1 contains the number of labels for each item purchased by the user; a4 is the weight for purchasing a commodity;
the position information comprises a plurality of labels set according to position scenes, the time information comprises a plurality of labels set according to time scenes, and each position information and each label at each time have corresponding weight; each recommendation information has the same basic recommendation score; the calculating, by combining the position information and the time, a recommendation score of the screened recommendation information according to a preset third algorithm specifically includes:
obtaining the screened labels of each piece of recommendation information, comparing whether the label of each piece of recommendation information is the same as the label under the position information of the user or not, and comparing whether the label of each piece of recommendation information is the same as the label under the time of the user or not;
if the label exists, multiplying the weight of the same label with the basic recommendation score of the recommendation information respectively, and taking the sum of the products as the basic recommendation score of the recommendation information.
2. The method of claim 1, wherein the pre-trained classifier employs a clustering algorithm to classify users into user groups.
3. The method according to claim 1, wherein the screening out the group recommendation labels of the user groups and the individual recommendation labels of the users in each user group according to a preset second algorithm specifically comprises:
screening a group recommendation label:
acquiring the number of labels contained in the recommendation information in the association data of each user in each user group, arranging the labels in a high-to-low order, and selecting the first J labels as the group recommendation labels of the user group;
screening personal recommendation labels:
and acquiring the number of each label contained in the recommendation information in the personal association data of the user, arranging the labels according to the sequence of the number from high to low, and selecting the first L labels as the personal recommendation labels of the user.
4. A recommendation information pushing apparatus, comprising: the device comprises a data establishing unit and an information pushing unit;
the data establishing unit specifically includes:
the database building module is used for building a recommendation information database, and each recommendation information comprises at least one label;
the associated data generation module is used for acquiring operation records of each user and respectively generating associated data of each user according to a preset first algorithm according to the operation records of each user; the associated data is the relation between the label included in each piece of recommendation information operated by the user and the operation record;
the user classification module is used for classifying the users into a plurality of user groups according to the associated data of the users and a classifier obtained by pre-training;
the recommendation label calculation module is used for respectively screening out the group recommendation labels of all user groups and the individual recommendation labels of all users in each user group according to the associated data of all users and a preset second algorithm;
the information pushing unit specifically includes:
the recommendation information screening module is used for acquiring the user group to which the user belongs, and screening recommendation information matched with the labels in the recommendation information database according to the group recommendation label of the user group to which the user belongs and the personal recommendation label of the user;
the recommendation information scoring module is used for acquiring position information of a user, determining time for pushing the recommendation information, calculating recommendation scores of the selected recommendation information according to a preset third algorithm by combining the position information and the time, arranging the recommendation information according to a sequence of the recommendation scores from high to low, and selecting the first K pieces of recommendation information to push to the user;
the recommendation information comprises commodities and contents; the operation records comprise purchase records of the commodities, praise records, comment records and forwarding records of the contents; in the associated data generation module, for each tag in the associated data, the preset first algorithm satisfies:
A1×lg(B1)+A2×lg(B2)+A3×lg(B3)+A4×lg(C1)
wherein B1 contains the number of tags for each user's approved content; a1 is a weight for agreeing on the content;
b2 contains the number of tags for each user-reviewed content; a2 is the weight for commenting on content;
b3 contains the number of the label for each user's forwarded content; a3 is the weight for forwarding the content;
c1 contains the number of labels for each item purchased by the user; a4 is the weight for purchasing a commodity;
the position information comprises a plurality of labels set according to position scenes, the time information comprises a plurality of labels set according to time scenes, and each position information and each label at each time have corresponding weight; each recommendation information has the same basic recommendation score;
the recommendation information scoring module is specifically configured to obtain each of the screened labels of the recommendation information, compare whether the label of each of the recommendation information is the same as the label of the user in the location information, and compare whether the label of each of the recommendation information is the same as the label of the user in the time;
if the label exists, multiplying the weight of the same label with the basic recommendation score of the recommendation information respectively, and taking the sum of the products as the basic recommendation score of the recommendation information.
5. The apparatus of claim 4, wherein the pre-trained classifier employs a clustering algorithm to classify users into a plurality of user groups in the user classification module.
6. The apparatus according to claim 4, wherein the recommended label calculation module specifically includes: the system comprises a group recommendation tag module and a personal recommendation tag module;
the group recommendation tags are used for acquiring the number of the tags contained in the recommendation information in the association data of each user in each user group, arranging the tags in a high-to-low order, and selecting the first J tags as the group recommendation tags of the user group;
the personal recommendation tags are used for acquiring the number of each tag contained in the recommendation information in the personal association data of the user, arranging the tags in a sequence from high to low, and selecting the first L tags as the personal recommendation tags of the user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910867360.4A CN110727859B (en) | 2019-09-12 | 2019-09-12 | Recommendation information pushing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910867360.4A CN110727859B (en) | 2019-09-12 | 2019-09-12 | Recommendation information pushing method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110727859A CN110727859A (en) | 2020-01-24 |
| CN110727859B true CN110727859B (en) | 2022-07-01 |
Family
ID=69218100
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910867360.4A Active CN110727859B (en) | 2019-09-12 | 2019-09-12 | Recommendation information pushing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110727859B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111083229B (en) * | 2019-12-24 | 2021-01-08 | 珠海大横琴科技发展有限公司 | Application information pushing method and device, electronic equipment and storage medium |
| CN113094577B (en) * | 2021-03-08 | 2024-03-15 | 北京达佳互联信息技术有限公司 | Information display method, related equipment and storage medium |
| CN113094578B (en) * | 2021-03-16 | 2023-03-31 | 平安普惠企业管理有限公司 | Deep learning-based content recommendation method, device, equipment and storage medium |
| CN113051476B (en) * | 2021-03-25 | 2023-06-13 | 北京百度网讯科技有限公司 | Method and device for sending message |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103399860A (en) * | 2013-07-04 | 2013-11-20 | 北京百纳威尔科技有限公司 | Content display method and device |
| CN103995839A (en) * | 2014-04-30 | 2014-08-20 | 兴天通讯技术（天津）有限公司 | Commodity recommendation optimizing method and system based on collaborative filtering |
| CN105468630A (en) * | 2014-09-04 | 2016-04-06 | 口碑控股有限公司 | Method and apparatus for providing user information |
| CN105872638A (en) * | 2016-04-19 | 2016-08-17 | 乐视控股（北京）有限公司 | Method and device for video push based on geographical position and server thereof |
| CN107688606A (en) * | 2017-07-26 | 2018-02-13 | 北京三快在线科技有限公司 | The acquisition methods and device of a kind of recommendation information, electronic equipment |
| CN108363821A (en) * | 2018-05-09 | 2018-08-03 | 深圳壹账通智能科技有限公司 | A kind of information-pushing method, device, terminal device and storage medium |
| CN109543111B (en) * | 2018-11-28 | 2021-09-21 | 广州虎牙信息科技有限公司 | Recommendation information screening method and device, storage medium and server |
-
2019
- 2019-09-12 CN CN201910867360.4A patent/CN110727859B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| 基于用户及群组特征的组推荐研究;郑伟;《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》;20190215;第I138-2627页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110727859A (en) | 2020-01-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110727859B (en) | Recommendation information pushing method and device | |
| CN107424043B (en) | Product recommendation method and device and electronic equipment | |
| CN109658206B (en) | Information recommendation method and device | |
| EP2304619B1 (en) | Correlated information recommendation | |
| CN103246980B (en) | Information output method and server | |
| CN112015998B (en) | Commodity recommendation method based on user portrait | |
| CN110175895B (en) | Article recommendation method and device | |
| CN105678317B (en) | Information processing method and server | |
| KR102000076B1 (en) | Method and server for recommending online sales channel on online shoppingmall intergrated management system | |
| WO2017092602A1 (en) | Method for screening information delivery user and server | |
| CN110750697B (en) | Merchant classification method, device, equipment and storage medium | |
| CN110941762B (en) | Method for improving sequencing result of personalized recommendation system | |
| CN103425664A (en) | Method and equipment for searching and displaying entity data units | |
| CN106294500A (en) | The method for pushing of content item, Apparatus and system | |
| KR20160131477A (en) | An e-commerce system based on interest category using related keywords | |
| CN108920647A (en) | Low-rank matrix based on spectral clustering fills TOP-N recommended method | |
| CN112417284A (en) | Method and device for pushing display information | |
| CN107093122B (en) | Object classification method and device | |
| KR101637986B1 (en) | System for providing the optimal products and method thereof | |
| CN111753180A (en) | Search method, search device, electronic equipment and computer storage medium | |
| CN110020918B (en) | Recommendation information generation method and system | |
| CN106326312B (en) | Information display method, device and system | |
| CN105704573A (en) | Television program shopping guide system and method thereof | |
| CN111177564A (en) | Product recommendation method and device | |
| CN112783468A (en) | Target object sorting method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |