CN110147497A - A kind of personalization content recommendation method towards younger population - Google Patents
A kind of personalization content recommendation method towards younger population Download PDFInfo
- Publication number
- CN110147497A CN110147497A CN201910405862.5A CN201910405862A CN110147497A CN 110147497 A CN110147497 A CN 110147497A CN 201910405862 A CN201910405862 A CN 201910405862A CN 110147497 A CN110147497 A CN 110147497A
- Authority
- CN
- China
- Prior art keywords
- user
- content
- training set
- recommendation
- discrete features
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Trade council of the present invention has opened a kind of personalization content recommendation method towards younger population, includes the following steps, S1, collects historical viewings behavior of the user to every recommendation, and as the training set of neural network model;S2, building neural network model;S3, training set is pre-processed, and it is included in pretreated training set is carried out in neural network model, it obtains and is included in as a result, obtained result of being included in does mean square error calculating, and carries out model training to neural network model as optimization aim to minimize mean square error result;S4, selection user, and its opposite content recommended carries out recommendation marking when to its recommendation.Advantage is: improving the degree of conformity of recommendation and age of user;Content exposure degree in addition to user's history interest, meeting age of user feature is improved, the teenager user visual field is widened;Under the premise of totally guaranteeing the degree of conformity of content and user interest, the history of overfitting user is avoided to like and formed information cocoon room.
Description
Technical field
The present invention relates to personalized recommendation algorithm fields more particularly to a kind of individualized content towards younger population to push away
Recommend method.
Background technique
With the fast development of information technology and network technology, there is explosive growth in global information, and mass data is presented
Before people, allow people while enjoying information resources abundant, also suffer from how to get it is actually useful to oneself
Part information.In face of this " data overrun " problem, there is search engine and both tools of recommended engine, help to manage
Solve the information requirement of user;The wherein user oriented dominant intention of search engine, i.e. user have specific access to information demand;And
Then user oriented recessive intention, i.e. user do not have specific access to information demand to recommended engine.Wherein recommended engine, especially
It is the recommended engine for having personalized recommendation function, the demand of user can be preferably solved, so personalized recommendation
The technology attention by more and more people in recent years, is increasingly becoming one of important need of content service provider.
It is pushed away because the user oriented recessive recessive intention for being intended to, thus how obtaining user of recommended engine becomes personalization
The emphasis of technical research is recommended, common means are all based on the historical behavior analysis user characteristics of user, and then form user's
Recessiveness is intended to.Traditional technical solution has three classes collaborative filtering: 1, based on the collaborative filtering of content, i.e., the recessiveness of user
It is intended to be equivalent to the feature for the thing liked before him;2, based on the collaborative filtering of user, i.e., the recessive of user is intended to equivalence
In other users similar with him so the thing liked;3, based on the collaborative filtering of matrix decomposition, i.e., the recessiveness of user is anticipated
Figure and the recessive character of content are solved to one group of hidden vector by matrix operation.The major defect of first two collaborative filtering exists
In, they can not cope with the sparse data scene of higher-dimension, and a kind of last collaborative filtering based on matrix decomposition, and because
Calculation amount is huge can not quickly to carry out model modification, and it can not handle the non-linear correlation between hidden feature, can not utilize
User's Figure Characteristics.
It is influenced by the collaborative filtering based on matrix decomposition, many new personalized recommendation technologies will all be thought in recent years
Road concentrates in the implicit vector expression of user's recessiveness intention, i.e., with the information requirement of one group of hidden vector expression user.In addition by
There is a collection of personalized recommendation technology based on neural network model, such as Factor minute in the development of machine learning techniques in recent years
Solution machine, Wide&Deep neural network etc.;But in algorithm design, specific consideration is done by unpromising teenager user group,
The surging of ctr is pursued simply and causes gradually narrowing for the visual field instead, and the information cocoon room formed for a long time can strangle teen-age creation
Property.
Summary of the invention
The purpose of the present invention is to provide a kind of personalization content recommendation methods towards younger population, to solve existing
There are foregoing problems present in technology.
To achieve the goals above, The technical solution adopted by the invention is as follows:
A kind of personalization content recommendation method towards younger population, includes the following steps,
S1, historical viewings behavior of the user to every recommendation is collected, and as the training of neural network model
Collection;
S2, building neural network model;
S3, training set is pre-processed, and be included in pretreated training set is carried out in neural network model, is obtained
It is included in as a result, obtained result of being included in is done mean square error calculating, and be optimization aim to mind to minimize mean square error result
Model training is carried out through network model;
S4, selection user, and its opposite content recommended carries out recommendation marking when to its recommendation.
Preferably, the training set being denoted as T, the T is expressed as follows,
T={ < X1,y1>, < X2,y2> ..., < XN,yN> }
Wherein, i=1,2 ..., N, N are the sum of behavioral data in training set, XiFor i-th of behavior number in training set
According to yiFor the feedback result of i-th of behavioral data in training set.
Preferably, according to user to the corresponding historical viewings behavior of every recommendation, determine that every recommendation is corresponding
YiValue;If user is one click behavior, y to the historical viewings behavior of recommendationi=1, if user is to recommendation
The historical viewings behavior of content is that single exposure does not click on behavior, then yi=0, if user is to the historical viewings row of recommendation
Behavior is not liked for a user's mark, then yi=-1.
Preferably, the XiIt is shown below,
Xi=(uidu,ageu,genderu,cated,typed,{tagd1,tagd2,...,tagdM});
Wherein, " uidu,ageu,genderu" for the main body of i-th of behavioral data of triggering, i.e., the feature of u-th user;"
cated,typed,{tagd1,tagd2,...,tagdMIt " is the object of i-th of behavioral data, i.e., the feature of the d articles content;uidu
It is the number of u-th of user, ageuIt is the age of u-th of user, genderuIt is the gender of u-th of user, catedIt is the d articles
The content type of content, typedIt is the ways of presentation of the d articles content, tagdjIt is j-th of label of the d articles content, j=1,
2 ..., M, M are the total number of labels of the d articles content.
Preferably, step S2 includes following content,
S201, using Customs Assigned Number as a monodrome discrete features, by an embeding layer, be converted to the numerical value of 32 dimensions to
Amount;
S202, using age of user as a monodrome discrete features, by an embeding layer, be converted to the numerical value of 64 dimensions to
Amount;
S203, using user's gender as a monodrome discrete features, by an embeding layer, be converted to the numerical value of 32 dimensions to
Amount;
S204, step S201, S202 are connected with numerical value vector acquired in S203, pass through the first full articulamentum, conversion
For the first numerical value vector of 128 dimensions;
S205, using content type as a monodrome discrete features, by an embeding layer, be converted to the numbers of 32 dimensions to
Amount;
S206, the number of 32 dimensions is converted to by an embeding layer using content ways of presentation as a monodrome discrete features
It is worth vector;
S207, the tag set by content, as a multivalue discrete features, by a sparse embeding layer, and will be more
The transformation result for being worth discrete features is added, and is converted to the numerical value vector of 64 dimensions;
S208, step S205, S206 are connected with numerical value vector obtained in S207, pass through the second full articulamentum, conversion
For the second value vector of 128 dimensions;
S209, the first numerical value vector sum second value vector is subjected to inner product operation, to obtain neural network model.
Preferably, step S3 includes following content,
S301, whole Customs Assigned Numbers is made into dictionary, the Customs Assigned Number for taking out the first behavioral data in training set exists
Index in dictionary obtains the monodrome discrete features of the Customs Assigned Number, and using the monodrome discrete features as the defeated of step S201
Enter;
S302, it is directed to teenager user group, limits the age range of user as 0 to 18, to first in training set
The age of user of behavioral data carries out special datum processing, obtains the monodrome discrete features of the age of user, and the monodrome is discrete
Input of the feature as step S202;
S303, user's gender is defined, including 0- is unknown, 1- male, 2- women, to first behavioral data in training set
User's gender quantize, obtain the monodrome discrete features of user's deformation, and as the input of step S203;
S304, it quantizes to the content type of first behavioral data in training set, obtains the content type
Monodrome discrete features, and using the monodrome discrete features as the input of step S205;
S305, it quantizes to the content ways of presentation of first behavioral data in training set, obtains the content exhibition
The monodrome discrete features of existing mode, and using the monodrome discrete features as the input of step S206;
S306, Hash is carried out using content tab set of the fnv32 hash algorithm to first behavioral data in training set
Change, obtains the multivalue discrete features of the content tab set, and using the multivalue discrete features as the input of step S207;
S307, the discrete features in above-mentioned steps are calculated using the neural network model obtained in step S209,
Obtain the calculated result of neural network model;First row in the calculated result and training set of the neural network model that will acquire
Difference is sought for the feedback result of data, and using the difference as the training error of first behavioral data;
All behavioral datas in S308, traversal training set are that a batch is trained with 256 datas, take one
The mean square error of batch is optimized as evaluation result, and to minimize mean square error result as optimization aim, to complete
The training of neural network model.
Preferably, when in step S4 to user's recommendation, the marking algorithm given a mark to recommendation is as follows,
Wherein, Score is final score;Divide based on BaseScore;M is neural network model, X'iIt is used for u-th
The behavioural characteristic that the feature at family and the feature of the d articles content are combined into after pretreatment;M(X'i) it is by X'iAs step S209
Input acquired in neural network model calculated result, Threshold is score threshold;A be model marking except power because
Son;timedIt is the time of the d articles content;B is the time except weight factor;Shuffle breaks up algorithm at random.
Preferably, as M (X'i) it is less than score threshold, then it is assumed that the d articles content is not liked by u-th of user, in this
The final score of appearance sets 0.
Preferably, described to break up algorithm centered on 1 at random, using C as random magnitude.
The beneficial effects of the present invention are: the personalization content recommendation method in the 1, present invention, use to user characteristics and
The method that content characteristic constructs multilayer neural network, and age differences are enhanced in modeling process, recommendation can be improved
With the degree of conformity of age of user.2, the recommended method in the present invention uses comprehensive marking algorithm, to model point when sequence
Drop power is carried out, random factor is added, improves the exposure of content in addition to user's history interest, meeting age of user feature
Degree, can widen the visual field of teenager user.3, it under the premise of totally guaranteeing the degree of conformity of content and user interest, avoids excessively
It is fitted the history hobby of user, avoids the formation of information cocoon room.
Detailed description of the invention
Fig. 1 is the flow chart of recommended method in the embodiment of the present invention;
Fig. 2 is the flow chart of recommendation marking algorithm in the embodiment of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing, to the present invention into
Row is further described.It should be appreciated that the specific embodiments described herein are only used to explain the present invention, it is not used to
Limit the present invention.
As shown in Fig. 1 to 2, the present invention provides a kind of personalization content recommendation methods towards younger population, including
Following steps,
S1, historical viewings behavior of the user to every recommendation is collected, and as the training of neural network model
Collect T;
S2, building neural network model M;
S3, training T is pre-processed, and is included in pretreated training set is carried out in neural network model M, obtained
It is included in as a result, obtained result of being included in is done mean square error calculating, and be optimization aim to mind to minimize mean square error result
Model training is carried out through network model;
S4, selection user, and its opposite content recommended carries out recommendation marking when to its recommendation.
As shown in Figure 1, the training set is denoted as T in the present embodiment, the T is expressed as follows,
T={ < X1,y1>, < X2,y2> ..., < XN,yN> }
Wherein, i=1,2 ..., N, N are the sum of behavioral data in training set, XiFor i-th of behavior number in training set
According to yiFor the feedback result of i-th of behavioral data in training set.
In the present embodiment, according to user to the corresponding historical viewings behavior of every recommendation, every recommendation is determined
Corresponding yiValue;If user is one click behavior, y to the historical viewings behavior of recommendationi=1, if user couple
The historical viewings behavior of recommendation is that single exposure does not click on behavior, then yi=0, if user is clear to the history of recommendation
User's mark of behavior of looking at does not like behavior, then yi=-1.
In the present embodiment, the XiIt is shown below,
Xi=(uidu,ageu,genderu,cated,typed,{tagd1,tagd2,...,tagdM});
Behavioral data XiIt is made of two parts feature, the main body including triggering this behavioral data, i.e. u-th user's
Feature (uidu,ageu,genderu) and this behavioral data object, i.e., the feature (cate of the d articles contentd,typed,
{tagd1,tagd2,...,tagdM});uiduIt is the number of u-th of user, ageuIt is the age of u-th of user, genderuIt is
The gender of u user, catedIt is the content type of the d articles content, typedIt is the ways of presentation of the d articles content, tagdjIt is d
J-th of label of content, j=1,2 ..., M, M are the total number of labels of the d articles content.
In the present embodiment, step S2 includes following content,
S201, using Customs Assigned Number as a monodrome discrete features, by an embeding layer, be converted to the numerical value of 32 dimensions to
Amount;
S202, using age of user as a monodrome discrete features, by an embeding layer, be converted to the numerical value of 64 dimensions to
Amount;
S203, using user's gender as a monodrome discrete features, by an embeding layer, be converted to the numerical value of 32 dimensions to
Amount;
S204, step S201, S202 are connected with numerical value vector acquired in S203, pass through the first full articulamentum, conversion
For the first numerical value vector of 128 dimensions;
S205, using content type as a monodrome discrete features, by an embeding layer, be converted to the numbers of 32 dimensions to
Amount;
S206, the number of 32 dimensions is converted to by an embeding layer using content ways of presentation as a monodrome discrete features
It is worth vector;
S207, the tag set by content, as a multivalue discrete features, by a sparse embeding layer, and will be more
The transformation result for being worth discrete features is added, and is converted to the numerical value vector of 64 dimensions;
S208, step S205, S206 are connected with numerical value vector obtained in S207, pass through the second full articulamentum, conversion
For the second value vector of 128 dimensions;
S209, the first numerical value vector sum second value vector is subjected to inner product operation, to obtain neural network model.
In the present embodiment, step S3 includes following content,
S301, whole Customs Assigned Numbers is made into dictionary, takes out the first behavioral data X in training set T1Customs Assigned Number
uid1Index in dictionary obtains the monodrome discrete features uid' of the Customs Assigned Number1, and using the monodrome discrete features as step
The input of rapid S201;
S302, it is directed to teenager user group, limits the age range of user as 0 to 18, to first in training set T
Behavioral data X1Age of user age1Special datum processing is carried out, the monodrome discrete features age' of the age of user is obtained1, and will
The input of the monodrome discrete features as step S202;
S303, user's gender is defined, including 0- is unknown, 1- male, 2- women, to first behavior number in training set T
According to X1User's gender gender1It quantizes, obtains the monodrome discrete features gender' of user's deformation1, and as
The input of step S203;
S304, to first behavioral data X in training set T1Content type cate1It quantizes, obtains the content
The monodrome discrete features cate' of classification1, and using the monodrome discrete features as the input of step S205;
S305, to the content ways of presentation type of first behavioral data in training set T1It quantizes, is somebody's turn to do
The monodrome discrete features type' of content ways of presentation1, and using the monodrome discrete features as the input of step S206;
S306, using fnv32 hash algorithm to the content tab set { tag of first behavioral data in training set T11,
tag12,...,tag1MHashed is carried out, obtain the multivalue discrete features { fnv32 (tag of the content tab set11),fnv32
(tag12),...,fnv32(tag1M), and using the multivalue discrete features as the input of step S207;
S307, the discrete features in above-mentioned steps are calculated using the neural network model obtained in step S209,
Obtain the calculated result of neural network model;The calculated result for the neural network model that will acquire and first in training set T
The feedback result y of behavioral data1Difference is sought, and using the difference as the training error of first behavioral data;
All behavioral datas in S308, traversal training set T are that a batch is trained with 256 datas, take one
The mean square error of batch is optimized as evaluation result, and to minimize mean square error result as optimization aim, optimization process
Using adam optimizer, to complete the training of neural network model.
In the present embodiment, as shown in Fig. 2, in step S4 to user's recommendation when, beat what recommendation was given a mark
Divide algorithm as follows,
Wherein, Score is final score;Divide based on BaseScore;M is neural network model, X'iIt is used for u-th
The behavioural characteristic that the feature at family and the feature of the d articles content are combined into after pretreatment;M(X'i) it is by X'iAs step S209
Input acquired in neural network model calculated result, Threshold is score threshold;A be model marking except power because
Son, purpose is first is that by M (X'i) it is reduced to the section of (- 1,1), second is that cutting down the influence of model marking, avoid overfitting user
History hobby;timedIt is the time of the d articles content;B is the time except weight factor, it is therefore an objective to be reduced to the influence of time
The section of (- 1,1);Shuffle breaks up algorithm at random.
In the present embodiment, as M (X'i) it is less than score threshold, then it is assumed that the d articles content is not liked by u-th of user, should
The final score of content sets 0.It is described to break up algorithm centered on 1 at random, using C as random magnitude, that is, random perturbation because
Son, to cut down the influence of model marking so that point there is a randomized jitter.
By using above-mentioned technical proposal disclosed by the invention, following beneficial effect has been obtained:
The present invention provides a kind of personalization content recommendation method towards younger population, use to user characteristics and
The method that content characteristic constructs multilayer neural network, and age differences are enhanced in modeling process, recommendation can be improved
With the degree of conformity of age of user;Meanwhile the recommended method in the present invention uses comprehensive marking algorithm, to model when sequence
Divide the drop power that carries out, random factor is added, improves the exposure of content in addition to user's history interest, meeting age of user feature
Luminosity can widen the visual field of teenager user;Under the premise of totally guaranteeing the degree of conformity of content and user interest, avoid excessively
It is fitted the history hobby of user, avoids the formation of information cocoon room.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
Depending on protection scope of the present invention.
Claims (9)
1. a kind of personalization content recommendation method towards younger population, it is characterised in that: include the following steps,
S1, historical viewings behavior of the user to every recommendation is collected, and as the training set of neural network model;
S2, building neural network model;
S3, training set is pre-processed, and be included in pretreated training set is carried out in neural network model, acquisition is included in
It as a result, obtained result of being included in is done mean square error calculating, and is optimization aim to nerve net to minimize mean square error result
Network model carries out model training;
S4, selection user, and its opposite content recommended carries out recommendation marking when to its recommendation.
2. the personalization content recommendation method according to claim 1 towards younger population, it is characterised in that: will be described
Training set is denoted as T, and the T is expressed as follows,
T={ < X1,y1>, < X2,y2> ..., < XN,yN> }
Wherein, i=1,2 ..., N, N are the sum of behavioral data in training set, XiFor i-th of behavioral data, y in training setiFor
The feedback result of i-th of behavioral data in training set.
3. the personalization content recommendation method according to claim 2 towards younger population, it is characterised in that: according to
Family determines the corresponding y of every recommendation to the corresponding historical viewings behavior of every recommendationiValue;If user is to pushing away
The historical viewings behavior for recommending content is one click behavior, then yi=1, if user is to the historical viewings behavior of recommendation
Single exposure does not click on behavior, then yi=0, if user does not like user's mark of historical viewings behavior of recommendation
Behavior, then yi=-1.
4. the personalization content recommendation method according to claim 2 towards younger population, it is characterised in that: the Xi
It is shown below,
Xi=(uidu,ageu,genderu,cated,typed,{tagd1,tagd2,...,tagdM});
Wherein, " uidu,ageu,genderu" for the main body of i-th of behavioral data of triggering, i.e., the feature of u-th user;"cated,
typed,{tagd1,tagd2,...,tagdMIt " is the object of i-th of behavioral data, i.e., the feature of the d articles content;uiduIt is u
The number of a user, ageuIt is the age of u-th of user, genderuIt is the gender of u-th of user, catedIt is the d articles content
Content type, typedIt is the ways of presentation of the d articles content, tagdjIt is j-th of label of the d articles content, j=1,2 ..., M, M
For the total number of labels of the d articles content.
5. the personalization content recommendation method according to claim 1 towards younger population, it is characterised in that: step S2
Including following content,
S201, the numerical value vector of 32 dimensions is converted to by an embeding layer using Customs Assigned Number as a monodrome discrete features;
S202, the numerical value vector of 64 dimensions is converted to by an embeding layer using age of user as a monodrome discrete features;
S203, the numerical value vector of 32 dimensions is converted to by an embeding layer using user's gender as a monodrome discrete features;
S204, step S201, S202 are connected with numerical value vector acquired in S203, by the first full articulamentum, are converted to 128
First numerical value vector of dimension;
S205, the digital vectors of 32 dimensions are converted to by an embeding layer using content type as a monodrome discrete features;
S206, using content ways of presentation as a monodrome discrete features, by an embeding layer, be converted to the numerical value of 32 dimensions to
Amount;
S207, the tag set by content, as a multivalue discrete features, by a sparse embeding layer, and by multivalue from
The transformation result for dissipating feature is added, and is converted to the numerical value vector of 64 dimensions;
S208, step S205, S206 are connected with numerical value vector obtained in S207, by the second full articulamentum, are converted to 128
The second value vector of dimension;
S209, the first numerical value vector sum second value vector is subjected to inner product operation, to obtain neural network model.
6. the personalization content recommendation method according to claim 1 towards younger population, it is characterised in that: step S3
Including following content,
S301, whole Customs Assigned Numbers is made into dictionary, takes out the Customs Assigned Number of the first behavioral data in training set in dictionary
In index, obtain the monodrome discrete features of the Customs Assigned Number, and using the monodrome discrete features as the input of step S201;
S302, it is directed to teenager user group, limits the age range of user as 0 to 18, to first behavior in training set
The age of user of data carries out special datum processing, obtains the monodrome discrete features of the age of user, and by the monodrome discrete features
Input as step S202;
S303, user's gender is defined, including 0- is unknown, 1- male, 2- women, to the use of first behavioral data in training set
Family gender quantizes, and obtains the monodrome discrete features of user's deformation, and as the input of step S203;
S304, it quantizes to the content type of first behavioral data in training set, obtains the monodrome of the content type
Discrete features, and using the monodrome discrete features as the input of step S205;
S305, it quantizes to the content ways of presentation of first behavioral data in training set, obtains the content side of showing
The monodrome discrete features of formula, and using the monodrome discrete features as the input of step S206;
S306, hashed is carried out using content tab set of the fnv32 hash algorithm to first behavioral data in training set, obtained
To the multivalue discrete features of the content tab set, and using the multivalue discrete features as the input of step S207;
S307, the discrete features in above-mentioned steps are calculated using the neural network model obtained in step S209, is obtained
The calculated result of neural network model;First behavior number in the calculated result and training set of the neural network model that will acquire
According to feedback result seek difference, and using the difference as the training error of first behavioral data;
All behavioral datas in S308, traversal training set are that a batch is trained with 256 datas, take a batch
Mean square error optimized as evaluation result, and to minimize mean square error result as optimization aim, to complete nerve
The training of network model.
7. the personalization content recommendation method according to claim 1 towards younger population, it is characterised in that: step S4
When the middle recommendation to user, the marking algorithm given a mark to recommendation is as follows,
Wherein, Score is final score;Divide based on BaseScore;M is neural network model, X'iFor u-th user's
The behavioural characteristic that the feature of feature and the d articles content is combined into after pretreatment;M(X'i) it is by X'iAs the defeated of step S209
Enter the calculated result of acquired neural network model, Threshold is score threshold;A is model marking except weight factor;
timedIt is the time of the d articles content;B is the time except weight factor;Shuffle breaks up algorithm at random.
8. the personalization content recommendation method according to claim 7 towards younger population, it is characterised in that: work as M
(X'i) it is less than score threshold, then it is assumed that the d articles content is not liked by u-th of user, and the final score of this content sets 0.
9. the personalization content recommendation method according to claim 7 towards younger population, it is characterised in that: it is described with
Machine breaks up algorithm centered on 1, using C as random magnitude.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910405862.5A CN110147497B (en) | 2019-05-15 | 2019-05-15 | Individual content recommendation method for teenager group |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910405862.5A CN110147497B (en) | 2019-05-15 | 2019-05-15 | Individual content recommendation method for teenager group |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110147497A true CN110147497A (en) | 2019-08-20 |
CN110147497B CN110147497B (en) | 2020-08-04 |
Family
ID=67595655
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910405862.5A Active CN110147497B (en) | 2019-05-15 | 2019-05-15 | Individual content recommendation method for teenager group |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110147497B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111612581A (en) * | 2020-05-18 | 2020-09-01 | 深圳市分期乐网络科技有限公司 | Method, device and equipment for recommending articles and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108304440A (en) * | 2017-11-01 | 2018-07-20 | 腾讯科技(深圳)有限公司 | Method, apparatus, computer equipment and the storage medium of game push |
CN108959603A (en) * | 2018-07-13 | 2018-12-07 | 北京印刷学院 | Personalized recommendation system and method based on deep neural network |
CN109145146A (en) * | 2018-09-07 | 2019-01-04 | 北京奇艺世纪科技有限公司 | A kind of data object recommended method, device and electronic equipment |
CN109241431A (en) * | 2018-09-07 | 2019-01-18 | 腾讯科技(深圳)有限公司 | A kind of resource recommendation method and device |
US20190114404A1 (en) * | 2017-10-18 | 2019-04-18 | Mastercard International Incorporated | Methods and systems for automatically configuring user authentication rules |
-
2019
- 2019-05-15 CN CN201910405862.5A patent/CN110147497B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190114404A1 (en) * | 2017-10-18 | 2019-04-18 | Mastercard International Incorporated | Methods and systems for automatically configuring user authentication rules |
CN108304440A (en) * | 2017-11-01 | 2018-07-20 | 腾讯科技(深圳)有限公司 | Method, apparatus, computer equipment and the storage medium of game push |
CN108959603A (en) * | 2018-07-13 | 2018-12-07 | 北京印刷学院 | Personalized recommendation system and method based on deep neural network |
CN109145146A (en) * | 2018-09-07 | 2019-01-04 | 北京奇艺世纪科技有限公司 | A kind of data object recommended method, device and electronic equipment |
CN109241431A (en) * | 2018-09-07 | 2019-01-18 | 腾讯科技(深圳)有限公司 | A kind of resource recommendation method and device |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111612581A (en) * | 2020-05-18 | 2020-09-01 | 深圳市分期乐网络科技有限公司 | Method, device and equipment for recommending articles and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN110147497B (en) | 2020-08-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104834686B (en) | A kind of video recommendation method based on mixing semantic matrix | |
CN103324665B (en) | Hot spot information extraction method and device based on micro-blog | |
CN106815297A (en) | A kind of academic resources recommendation service system and method | |
CN105045875B (en) | Personalized search and device | |
CN112214685A (en) | Knowledge graph-based personalized recommendation method | |
CN103064903B (en) | Picture retrieval method and device | |
CN107122455A (en) | A kind of network user's enhancing method for expressing based on microblogging | |
CN104899273A (en) | Personalized webpage recommendation method based on topic and relative entropy | |
CN111143672B (en) | Knowledge graph-based professional speciality scholars recommendation method | |
CN110188346A (en) | A kind of network security bill part intelligence analysis method based on information extraction | |
CN112966091B (en) | Knowledge map recommendation system fusing entity information and heat | |
CN111488524B (en) | Attention-oriented semantic-sensitive label recommendation method | |
CN109992674B (en) | Recommendation method fusing automatic encoder and knowledge graph semantic information | |
CN113806630B (en) | Attention-based multi-view feature fusion cross-domain recommendation method and device | |
Liu et al. | Using collaborative filtering algorithms combined with Doc2Vec for movie recommendation | |
CN110134792A (en) | Text recognition method, device, electronic equipment and storage medium | |
CN109902229A (en) | A kind of interpretable recommended method based on comment | |
CN110222172A (en) | A kind of multi-source network public sentiment Topics Crawling method based on improvement hierarchical clustering | |
CN111949848A (en) | Cross-platform propagation situation assessment and grading method based on specific events | |
CN114781503A (en) | Click rate estimation method based on depth feature fusion | |
CN111339429B (en) | Information recommendation method | |
CN110968675B (en) | Recommendation method and system based on multi-field semantic fusion | |
CN110147497A (en) | A kind of personalization content recommendation method towards younger population | |
Ravanifard et al. | Content-aware listwise collaborative filtering | |
CN102866997B (en) | The treating method and apparatus of user data |
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 |