CN110147497B - Individual content recommendation method for teenager group - Google Patents
Individual content recommendation method for teenager group Download PDFInfo
- Publication number
- CN110147497B CN110147497B CN201910405862.5A CN201910405862A CN110147497B CN 110147497 B CN110147497 B CN 110147497B CN 201910405862 A CN201910405862 A CN 201910405862A CN 110147497 B CN110147497 B CN 110147497B
- Authority
- CN
- China
- Prior art keywords
- user
- content
- training set
- neural network
- taking
- 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
- 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
The invention discloses a personalized content recommendation method facing teenager groups, which comprises the following steps of S1, collecting historical browsing behaviors of each piece of recommended content by a user, and using the historical browsing behaviors as a training set of a neural network model; s2, constructing a neural network model; s3, preprocessing the training set, putting the preprocessed training set into a neural network model, obtaining an inclusion result, performing mean square error calculation on the obtained inclusion result, and performing model training on the neural network model by taking the minimum mean square error result as an optimization target; and S4, selecting the user, and scoring the recommendation of the recommended content when the content is recommended to the user. The advantages are that: the conformity between the recommended content and the age of the user is improved; the content exposure degree which is in accordance with the age characteristics of the user except the historical interest of the user is improved, and the vision field of the teenager user is widened; on the premise of generally ensuring the conformity of the content and the user interest, the history preference of the user is prevented from being excessively fitted and an information cocoon house is prevented from being formed.
Description
Technical Field
The invention relates to the field of personalized recommendation algorithms, in particular to a juvenile group-oriented personalized content recommendation method.
Background
With the rapid development of information technology and network technology, global information has explosively increased, and mass data is presented to people, so that people can enjoy abundant information resources and simultaneously trouble how to obtain the information which is really useful for themselves. In the face of the problem of data overload, two tools, namely a search engine and a recommendation engine, appear to help understand the information requirements of users; the search engine faces the explicit intention of the user, namely the user has clear information acquisition requirements; the recommendation engine faces the implicit intention of the user, that is, the user does not have an explicit information acquisition requirement. The recommendation engine, especially the recommendation engine with personalized recommendation function, can better solve the implicit information requirement of the user, so that the personalized recommendation technology is more and more emphasized by more people in recent years, and is gradually one of the important requirements of content service providers.
Because the recommendation engine faces the implicit intention of the user, how to acquire the implicit intention of the user becomes the key point of personalized recommendation technology research, and the common means is to analyze the characteristics of the user based on the historical behavior of the user so as to form the implicit intention of the user. The traditional technical scheme comprises three types of collaborative filtering technologies: 1. content-based collaborative filtering, i.e., equating the user's implicit intent to the characteristics of what he previously liked; 2. collaborative filtering based on a user, i.e. equating the implicit intent of the user to something similar to him that other users like; 3. and (3) collaborative filtering based on matrix decomposition, namely solving the implicit intention of the user and the implicit characteristics of the content into a group of implicit vectors through matrix operation. The first two collaborative filtering techniques have the main disadvantages that they cannot cope with high-dimensional sparse data scenes, and the last collaborative filtering technique based on matrix decomposition cannot rapidly update the model due to huge calculation amount, cannot process the nonlinear association between hidden features, and cannot utilize the user portrait features.
Under the influence of a collaborative filtering technology based on matrix decomposition, many new personalized recommendation technologies in recent years focus on the idea of implicit vector expression of implicit intentions of users, namely, a group of implicit vectors is used for expressing the information requirements of the users. In addition, due to the development of machine learning technology in recent years, a batch of personalized recommendation technology based on neural network models appears, such as a factorization machine, Wide & Deep neural networks and the like; however, in terms of algorithm design, no specific consideration is made for the teenager user group, the vision field is gradually narrowed due to the fact that the ctr is increased once, and the creativity of the teenagers can be killed due to the long-term formation of the information cocoon house.
Disclosure of Invention
The present invention aims to provide a method for recommending personalized content for teenager groups, so as to solve the aforementioned problems in the prior art.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a personalized content recommendation method facing teenager groups comprises the following steps,
s1, collecting historical browsing behaviors of each piece of recommended content by the user, and taking the historical browsing behaviors as a training set of a neural network model;
s2, constructing a neural network model;
s3, preprocessing the training set, putting the preprocessed training set into a neural network model, obtaining an inclusion result, performing mean square error calculation on the obtained inclusion result, and performing model training on the neural network model by taking the minimum mean square error result as an optimization target;
and S4, selecting the user, and scoring the recommendation of the recommended content when the content is recommended to the user.
Preferably, the training set is denoted as T, which is expressed as follows,
T＝{＜X1,y1＞,＜X2,y2＞,...,＜XN,yN＞}
where, i is 1, 2., N is the total number of behavior data in the training set, XiFor the ith behavioral data, y, in the training setiAnd feeding back the result of the ith behavior data in the training set.
Preferably, y corresponding to each piece of recommended content is determined according to the historical browsing behavior of the user corresponding to each piece of recommended contentiA value of (d); y if the user's historical browsing behavior for recommended content is one-click behaviori1, if the user's historical browsing behavior for recommended content is an exposure-to-no-click behavior, yi0, if the user marks dislike behavior once for the history browsing behavior of the recommended content, yi＝-1。
Preferably, X isiAs shown in the following formula,
Xi＝(uidu,ageu,genderu,cated,typed,{tagd1,tagd2,...,tagdM})；
wherein, the uidu,ageu,genderu"is the main body triggering the ith behavior data, namely the characteristics of the u-th user; "cated,typed,{tagd1,tagd2,...,tagdMThe item is the object of the ith behavior data, namely the characteristic of the d-th item of content; uiduIs the number of the u-th user, ageuIs the age, gender, of the u-th useruIs the gender, cat, of the u-th userdIs the content type of the d-th contentdIs the presentation mode, tag, of the d-th contentdjIs the jth label of the d-th content, j is 1, 2.
Preferably, step S2 includes the following,
s201, converting a user number serving as a single-value discrete feature into a 32-dimensional numerical vector through an embedded layer;
s202, converting the age of the user into a 64-dimensional numerical vector through an embedding layer as a single-value discrete feature;
s203, converting the gender of the user into a 32-dimensional numerical vector through an embedded layer as a single-value discrete feature;
s204, connecting the numerical vectors obtained in the steps S201, S202 and S203, and converting the numerical vectors into a 128-dimensional first numerical vector through a first full-connection layer;
s205, converting the content category as a single-value discrete feature into a 32-dimensional digital vector through an embedding layer;
s206, converting the content display mode serving as a single-value discrete feature into a 32-dimensional numerical vector through an embedded layer;
s207, taking a label set of the content as a multi-valued discrete feature, passing through a sparse embedding layer, adding conversion results of the multi-valued discrete feature, and converting the added conversion results into a 64-dimensional numerical vector;
s208, connecting the numerical vectors obtained in the steps S205, S206 and S207, and converting the numerical vectors into 128-dimensional second numerical vectors through a second full-connection layer;
s209, carrying out inner product operation on the first numerical value vector and the second numerical value vector, thereby obtaining the neural network model.
Preferably, step S3 includes the following,
s301, making all the user numbers into a dictionary, taking out the indexes of the user numbers of the first behavior data in the training set in the dictionary to obtain the single-value discrete features of the user numbers, and taking the single-value discrete features as the input of the step S201;
s302, aiming at a teenager user group, limiting the age interval of the user to be 0-18, performing special value processing on the user age of the first behavior data in the training set to obtain a single-value discrete feature of the user age, and taking the single-value discrete feature as the input of the step S202;
s303, defining user genders including 0-unknown, 1-male and 2-female, digitizing the user gender of the first behavior data in the training set to obtain a single-value discrete characteristic of user deformation, and taking the single-value discrete characteristic as the input of the step S203;
s304, digitizing the content type of the first behavior data in the training set to obtain a single-value discrete feature of the content type, and taking the single-value discrete feature as the input of the step S205;
s305, digitizing the content presentation mode of the first behavior data in the training set to obtain a single-value discrete feature of the content presentation mode, and taking the single-value discrete feature as the input of the step S206;
s306, hashing the content label set of the first behavior data in the training set by adopting an fnv32 hashing algorithm to obtain a multi-value discrete feature of the content label set, and taking the multi-value discrete feature as the input of the step S207;
s307, calculating the discrete features in the step by adopting the neural network model obtained in the step S209 to obtain a calculation result of the neural network model; calculating a difference value between the calculation result of the obtained neural network model and a feedback result of the first behavior data in the training set, and taking the difference value as a training error of the first behavior data;
and S308, traversing all the behavior data in the training set, training by taking 256 pieces of data as one batch, taking the mean square error of one batch as a judgment result, and optimizing by taking the result of the minimized mean square error as an optimization target, thereby completing the training of the neural network model.
Preferably, when recommending content to the user in step S4, the scoring algorithm for scoring the recommended content is as follows,
wherein Score is the final Score; base score; m is a neural netVein model, X'iThe characteristics of the u-th user and the characteristics of the d-th content are combined into behavior characteristics after being preprocessed; m (X'i) Is prepared from X'iThreshold is a score Threshold as a calculation result of the neural network model acquired as an input of step S209; a is the dividing factor of the model score; timedIs the time of the d-th piece of content; b is a weight division factor of time; shuffle random scatter algorithm.
Preferably, when M (X'i) If the score is less than the score threshold, the ith content is considered to be not liked by the u-th user, and the final score of the piece of content is set to 0.
Preferably, the random break-up algorithm is centered at 1 and has a random amplitude of C.
The invention has the beneficial effects that: 1. the personalized content recommendation method adopts a method for constructing a multilayer neural network for the user characteristics and the content characteristics, strengthens the age difference in the modeling process, and can improve the conformity between the recommended content and the age of the user. 2. The recommendation method adopts a comprehensive scoring algorithm, reduces the weight of the model scores during sorting, adds random factors, improves the exposure of the content which is in line with the age characteristics of the user except the historical interest of the user, and can widen the visual field of teenager users. 3. On the premise of ensuring the conformity of the content and the user interest in the whole, the history preference of the user is prevented from being excessively fitted, and the formation of an information cocoon house is prevented.
Drawings
FIG. 1 is a flow chart of a recommendation method in an embodiment of the invention;
fig. 2 is a flowchart of a recommended content scoring algorithm in an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings. It should be understood that the detailed description and specific examples, while indicating the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.
As shown in fig. 1 to 2, the present invention provides a method for recommending personalized contents to a teenager group, comprising the steps of,
s1, collecting historical browsing behaviors of each piece of recommended content by the user, and using the historical browsing behaviors as a training set T of the neural network model;
s2, constructing a neural network model M;
s3, preprocessing the training T, storing the preprocessed training set into a neural network model M, acquiring an inclusion result, performing mean square error calculation on the obtained inclusion result, and performing model training on the neural network model by taking a minimum mean square error result as an optimization target;
and S4, selecting the user, and scoring the recommendation of the recommended content when the content is recommended to the user.
As shown in fig. 1, in this embodiment, the training set is denoted as T, which is expressed as follows,
T＝{＜X1,y1＞,＜X2,y2＞,...,＜XN,yN＞}
where, i is 1, 2., N is the total number of behavior data in the training set, XiFor the ith behavioral data, y, in the training setiAnd feeding back the result of the ith behavior data in the training set.
In this embodiment, y corresponding to each piece of recommended content is determined according to the historical browsing behavior of the user corresponding to each piece of recommended contentiA value of (d); y if the user's historical browsing behavior for recommended content is one-click behaviori1, if the user's historical browsing behavior for recommended content is an exposure-to-no-click behavior, yi0, if the user marks dislike behavior once for the history browsing behavior of the recommended content, yi＝-1。
In this example, X isiAs shown in the following formula,
Xi＝(uidu,ageu,genderu,cated,typed,{tagd1,tagd2,...,tagdM})；
behavior data XiIs composed of two-part features including a main body for triggering the behavior dataI.e. the characteristics (uid) of the u-th useru,ageu,genderu) And the object of this behavior data, i.e., the feature (cat) of the d-th contentd,typed,{tagd1,tagd2,...,tagdM})；uiduIs the number of the u-th user, ageuIs the age, gender, of the u-th useruIs the gender, cat, of the u-th userdIs the content type of the d-th contentdIs the presentation mode, tag, of the d-th contentdjIs the jth label of the d-th content, j is 1, 2.
In this embodiment, step S2 includes the following steps,
s201, converting a user number serving as a single-value discrete feature into a 32-dimensional numerical vector through an embedded layer;
s202, converting the age of the user into a 64-dimensional numerical vector through an embedding layer as a single-value discrete feature;
s203, converting the gender of the user into a 32-dimensional numerical vector through an embedded layer as a single-value discrete feature;
s204, connecting the numerical vectors obtained in the steps S201, S202 and S203, and converting the numerical vectors into a 128-dimensional first numerical vector through a first full-connection layer;
s205, converting the content category as a single-value discrete feature into a 32-dimensional digital vector through an embedding layer;
s206, converting the content display mode serving as a single-value discrete feature into a 32-dimensional numerical vector through an embedded layer;
s207, taking a label set of the content as a multi-valued discrete feature, passing through a sparse embedding layer, adding conversion results of the multi-valued discrete feature, and converting the added conversion results into a 64-dimensional numerical vector;
s208, connecting the numerical vectors obtained in the steps S205, S206 and S207, and converting the numerical vectors into 128-dimensional second numerical vectors through a second full-connection layer;
s209, carrying out inner product operation on the first numerical value vector and the second numerical value vector, thereby obtaining the neural network model.
In this embodiment, step S3 includes the following steps,
s301, making all user numbers into a dictionary, and taking out first behavior data X in a training set T1User number uid1Obtaining the single-valued discrete feature uid 'of the user number by indexing in a dictionary'1And the single-valued discrete characteristic is taken as the input of the step S201;
s302, aiming at a teenager user group, limiting the age interval of the users to be 0 to 18, and aiming at first behavior data X in a training set T1Age of the user1Performing special value processing to obtain a single value discrete feature age 'of the user age'1And the single-valued discrete characteristic is taken as the input of the step S202;
s303, defining the gender of the user, including 0-unknown, 1-male and 2-female, and comparing the first behavior data X in the training set T1User gender generator1Digitizing to obtain single value discrete feature generator of user deformation'1And takes it as the input of step S203;
s304, comparing the first behavior data X in the training set T1Content category of1Digitizing to obtain a single value discrete feature cat 'of the content category'1And the single-valued discrete feature is taken as the input of step S205;
s305, displaying the content of the first behavior data in the training set T in a mode type1Digitizing to obtain single value discrete feature type of the content display mode'1And the single-valued discrete feature is taken as the input of step S206;
s306, adopting fnv32 Hash algorithm to carry out content label set { tag) on the first behavior data in the training set T11,tag12,...,tag1MHash to obtain multiple-valued discrete characteristics (fnv 32 (tag)) of the content tag set11),fnv32(tag12),...,fnv32(tag1M) And taking the multivalued discrete characteristic as the input of step S207;
s307, adopting the neural network model obtained in the step S209Calculating the discrete characteristics in the steps to obtain a calculation result of the neural network model; the obtained calculation result of the neural network model and the feedback result y of the first behavior data in the training set T are compared1Calculating a difference value, and taking the difference value as a training error of the first behavior data;
s308, traversing all the behavior data in the training set T, training by taking 256 pieces of data as one batch, taking the mean square error of one batch as a judgment result, and optimizing by taking the result of the minimized mean square error as an optimization target, wherein an adam optimizer is adopted in the optimization process, so that the training of the neural network model is completed.
In this embodiment, as shown in fig. 2, when recommending content to the user in step S4, the scoring algorithm for scoring the recommended content is as follows,
wherein Score is the final Score; base score; m is the neural network model, X'iThe characteristics of the u-th user and the characteristics of the d-th content are combined into behavior characteristics after being preprocessed; m (X'i) Is prepared from X'iThreshold is a score Threshold as a calculation result of the neural network model acquired as an input of step S209; a is a weight-dividing factor for model scoring, with the objective of dividing M (X'i) The range is reduced to (-1,1), the influence of model scoring is reduced, and the history preference of a user is prevented from being over-fitted; timedIs the time of the d-th piece of content; b is a time division factor, aiming to reduce the influence of time to the interval of (-1, 1); shuffle random scatter algorithm.
In this example, when M (X'i) If the score is less than the score threshold, the ith content is considered to be not liked by the u-th user, and the final score of the piece of content is set to 0. The random scattering algorithm takes 1 as the center and C as the random amplitude, namely the random disturbance factor, so that the score has random jitter, and the influence of model scoring is reduced.
By adopting the technical scheme disclosed by the invention, the following beneficial effects are obtained:
the invention provides a personalized content recommendation method facing adolescent groups, which adopts a method of constructing a multilayer neural network for user characteristics and content characteristics, strengthens age difference in a modeling process and can improve the conformity between recommended content and user age; meanwhile, the recommendation method adopts a comprehensive scoring algorithm, reduces the weight of the model scores during sorting, adds random factors, improves the exposure of the content which is in line with the age characteristics of the user except the historical interest of the user, and can widen the visual field of teenager users; on the premise of ensuring the conformity of the content and the user interest in the whole, the history preference of the user is prevented from being excessively fitted, and the formation of an information cocoon house is prevented.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and improvements can be made without departing from the principle of the present invention, and such modifications and improvements should also be considered within the scope of the present invention.
Claims (3)
1. A personalized content recommendation method for teenager groups is characterized in that: comprises the following steps of (a) carrying out,
s1, collecting historical browsing behaviors of each piece of recommended content by the user, and taking the historical browsing behaviors as a training set of a neural network model;
s2, constructing a neural network model; meanwhile, hidden language meaning vectorization modeling is carried out by combining the user number, the user age, the user gender and the attributes of the user historical browsing recommended content to serve as the interest vector of the user;
s3, preprocessing the training set, putting the preprocessed training set into a neural network model, obtaining an inclusion result, performing mean square error calculation on the obtained inclusion result, and performing model training on the neural network model by taking the minimum mean square error result as an optimization target;
s4, selecting a user, and when recommending content to the user, scoring the recommendation of the content to the user;
the step S2 includes the following contents,
s201, converting a user number serving as a single-value discrete feature into a 32-dimensional numerical vector through an embedded layer;
s202, converting the age of the user into a 64-dimensional numerical vector through an embedding layer as a single-value discrete feature;
s203, converting the gender of the user into a 32-dimensional numerical vector through an embedded layer as a single-value discrete feature;
s204, connecting the numerical vectors obtained in the steps S201, S202 and S203, and converting the numerical vectors into a 128-dimensional first numerical vector through a first full-connection layer;
s205, converting the content category as a single-value discrete feature into a 32-dimensional digital vector through an embedding layer;
s206, converting the content display mode serving as a single-value discrete feature into a 32-dimensional numerical vector through an embedded layer;
s207, taking a label set of the content as a multi-valued discrete feature, passing through a sparse embedding layer, adding conversion results of the multi-valued discrete feature, and converting the added conversion results into a 64-dimensional numerical vector;
s208, connecting the numerical vectors obtained in the steps S205, S206 and S207, and converting the numerical vectors into 128-dimensional second numerical vectors through a second full-connection layer;
s209, carrying out inner product operation on the first numerical vector and the second numerical vector to obtain a neural network model;
when recommending content to the user in step S4, the scoring algorithm for scoring the recommended content is as follows,
wherein Score is the final Score; base score; m is the neural network model, X'iThe characteristics of the u-th user and the characteristics of the d-th content are combined into behavior characteristics after being preprocessed; m (X'i) Is prepared from X'iTh as a calculation result of the neural network model acquired as an input of step S209Threshold is the score threshold; a is the dividing factor of the model score; timedIs the time of the d-th piece of content; b is a weight division factor of time; shuffle random break-up algorithm; the random break-up algorithm takes 1 as the center and C as the random amplitude;
let the training set be denoted as T, which is expressed as follows,
T＝{<X1,y1>,<X2,y2>,...,<XN,yN>}
where, i is 1, 2., N is the total number of behavior data in the training set, XiFor the ith behavioral data, y, in the training setiFeedback results for the ith behavior data in the training set;
said XiAs shown in the following formula,
Xi＝(uidu,ageu,genderu,cated,typed,{tagd1,tagd2,...,tagdM})；
wherein, the uidu,ageu,genderu"is the main body triggering the ith behavior data, namely the characteristics of the u-th user; "cated,typed,{tagd1,tagd2,...,tagdMThe item is the object of the ith behavior data, namely the characteristic of the d-th item of content; uiduIs the number of the u-th user, ageuIs the age, gender, of the u-th useruIs the gender, cat, of the u-th userdIs the content type of the d-th contentdIs the presentation mode, tag, of the d-th contentdjJ is the jth label of the d-th content, wherein j is 1,2, and M is the total number of labels of the d-th content;
the step S3 includes the following contents,
s301, making all the user numbers into a dictionary, taking out the indexes of the user numbers of the first behavior data in the training set in the dictionary to obtain the single-value discrete features of the user numbers, and taking the single-value discrete features as the input of the step S201;
s302, aiming at a teenager user group, limiting the age interval of the user to be 0-18, performing special value processing on the user age of the first behavior data in the training set to obtain a single-value discrete feature of the user age, and taking the single-value discrete feature as the input of the step S202;
s303, defining user genders including 0-unknown, 1-male and 2-female, digitizing the user gender of the first behavior data in the training set to obtain a single-value discrete characteristic of user deformation, and taking the single-value discrete characteristic as the input of the step S203;
s304, digitizing the content type of the first behavior data in the training set to obtain a single-value discrete feature of the content type, and taking the single-value discrete feature as the input of the step S205;
s305, digitizing the content presentation mode of the first behavior data in the training set to obtain a single-value discrete feature of the content presentation mode, and taking the single-value discrete feature as the input of the step S206;
s306, hashing the content label set of the first behavior data in the training set by adopting an fnv32 hashing algorithm to obtain a multi-value discrete feature of the content label set, and taking the multi-value discrete feature as the input of the step S207;
s307, calculating the discrete features in the step by adopting the neural network model obtained in the step S209 to obtain a calculation result of the neural network model; calculating a difference value between the calculation result of the obtained neural network model and a feedback result of the first behavior data in the training set, and taking the difference value as a training error of the first behavior data;
and S308, traversing all the behavior data in the training set, training by taking 256 pieces of data as one batch, taking the mean square error of one batch as a judgment result, and optimizing by taking the result of the minimized mean square error as an optimization target, thereby completing the training of the neural network model.
2. The method of claim 1, wherein the method comprises: according to the historical browsing behavior of the user corresponding to each piece of recommended content, y corresponding to each piece of recommended content is determinediA value of (d); if the user is about the recommended contentIf the historical browsing behavior is one-click behavior, then yi1, if the user's historical browsing behavior for recommended content is an exposure-to-no-click behavior, yi0, if the user marks dislike behavior once for the history browsing behavior of the recommended content, yi＝-1。
3. The method of claim 1, wherein the method comprises: when M (X'i) If the score is less than the score threshold, the ith content is considered to be not liked by the u-th user, and the final score of the piece of content is set to 0.
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 CN110147497A (en) | 2019-08-20 |
| CN110147497B true 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) |
Families Citing this family (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 |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10650128B2 (en) * | 2017-10-18 | 2020-05-12 | Mastercard International Incorporated | Methods and systems for automatically configuring user authentication rules |
| CN108304440B (en) * | 2017-11-01 | 2021-08-31 | 腾讯科技（深圳）有限公司 | Game pushing method and device, computer equipment and storage medium |
| CN108959603B (en) * | 2018-07-13 | 2022-03-29 | 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 |
| CN109241431B (en) * | 2018-09-07 | 2023-11-07 | 腾讯科技（深圳）有限公司 | Resource recommendation method and device |
-
2019
- 2019-05-15 CN CN201910405862.5A patent/CN110147497B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110147497A (en) | 2019-08-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110298037B (en) | Convolutional neural network matching text recognition method based on enhanced attention mechanism | |
| CN106021364B (en) | Foundation, image searching method and the device of picture searching dependency prediction model | |
| CN110162593A (en) | A kind of processing of search result, similarity model training method and device | |
| CN110532379B (en) | Electronic information recommendation method based on LSTM (least Square TM) user comment sentiment analysis | |
| CN107729488A (en) | A kind of information recommendation method and device | |
| CN101447020B (en) | Pornographic image recognizing method based on intuitionistic fuzzy | |
| CN105045875B (en) | Personalized search and device | |
| CN112613552B (en) | Convolutional neural network emotion image classification method combined with emotion type attention loss | |
| CN107122455A (en) | A kind of network user's enhancing method for expressing based on microblogging | |
| CN111898031A (en) | Method and device for obtaining user portrait | |
| CN103678618A (en) | Web service recommendation method based on socializing network platform | |
| CN110110225B (en) | Online education recommendation model based on user behavior data analysis and construction method | |
| CN110457514A (en) | A kind of multi-tag image search method based on depth Hash | |
| CN106951471A (en) | A kind of construction method of the label prediction of the development trend model based on SVM | |
| CN109753602A (en) | A kind of across social network user personal identification method and system based on machine learning | |
| CN110990670B (en) | Growth incentive book recommendation method and recommendation system | |
| CN112749330B (en) | Information pushing method, device, computer equipment and storage medium | |
| CN112801760A (en) | Sequencing optimization method and system of content personalized recommendation system | |
| CN114693397A (en) | Multi-view multi-modal commodity recommendation method based on attention neural network | |
| CN108647800A (en) | A kind of online social network user missing attribute forecast method based on node insertion | |
| CN113239159A (en) | Cross-modal retrieval method of videos and texts based on relational inference network | |
| CN115840853A (en) | Course recommendation system based on knowledge graph and attention network | |
| CN112529638A (en) | Service demand dynamic prediction method and system based on user classification and deep learning | |
| CN112258250A (en) | Target user identification method and device based on network hotspot and computer equipment | |
| CN116935170A (en) | Processing method and device of video processing model, computer equipment and storage medium |
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 |