CN111259266B - Internet content recommendation method and system - Google Patents
Internet content recommendation method and system Download PDFInfo
- Publication number
- CN111259266B CN111259266B CN202010089194.2A CN202010089194A CN111259266B CN 111259266 B CN111259266 B CN 111259266B CN 202010089194 A CN202010089194 A CN 202010089194A CN 111259266 B CN111259266 B CN 111259266B
- Authority
- CN
- China
- Prior art keywords
- item type
- leader
- candidate
- weight
- user
- 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
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/10—Office automation; Time management
- G06Q10/105—Human resources
- G06Q10/1053—Employment or hiring
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Strategic Management (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- General Engineering & Computer Science (AREA)
- Economics (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to an internet content recommendation method and system, wherein the method comprises the following steps: selecting k from n known users as candidate guides, and initializing the weight of each candidate guide related to the type of the item; training the weight of each candidate leader related to the item type to update the weight of each candidate leader related to the item type, and selecting the candidate leader with the highest weight corresponding to each updated item type as a leader; and when a new user performs interviewing, recommending the content of the new user through the leader corresponding to the item type selected by the new user. According to the scheme, the problem of cold start of the recommendation system is solved on the basis of not needing metadata of a new user.
Description
Technical Field
The invention relates to the field of internet information, in particular to an internet content recommendation method and system.
Background
With the advent of the big data age, a great amount of various multimedia contents such as texts, pictures, audios and videos are accumulated on the internet. With the rapid popularization of social networks (e.g., facebook, twitter, newcastle of new waves, weChat friend circle, etc.) and devices supporting wireless data access (smartphones and tablet computers), people can freely create, upload, and share various types of multimedia content anytime and anywhere.
This has led to the explosion of the volume of data coming back to the internet already carrying large amounts of data; on one hand, extremely rich internet content can meet the personalized requirements of each user; on the other hand, the massive internet data also makes it difficult for users to quickly and accurately find information of interest; as a distributor of internet content, it is difficult to make the content stand out from mass data and accurately distribute the content to a target user group; this problem, known as "information overload," has become particularly acute in the current internet era.
One very potential solution to the information overload problem is the recommendation system; the recommendation system is a tool for helping a user to quickly find useful information, is different from a search engine, does not need the user to provide clear requirements, and is a personalized information recommendation system for recommending information, products and the like which are interested by the user to the user according to the information requirements, interests and the like of the user; compared with a search engine, the recommendation system carries out personalized calculation by researching the interest preference of the user, and the system finds the interest points of the user, thereby guiding the user to find the own information requirement; a good recommendation system not only can provide personalized services for users, but also can establish close relation with the users to enable the users to generate dependence on the recommendation; recommendation systems are now widely used in many fields, the most typical of which with good development and application prospects is the field of e-commerce; meanwhile, the research heat of the academic community on the recommendation system is high all the time, and an independent subject is gradually formed.
However, recommendation systems face many challenges, such as new user cold start issues; the problem of cold start of a new user means that for a recommendation system, the system cannot recommend items meeting the interest preference of the new user because the system does not know the interest preference of the new user due to lack of historical interaction data of the user or even no historical interaction data of the user; if a new user does not get a good experience when entering the system for the first time, the new user may be refused to use again after a repulsive mind is generated, and therefore, the problem of cold start of the new user is to be solved urgently.
At present, the research idea of many scholars at home and abroad aiming at the cold start problem is to embed a new user or a new project into the existing model by utilizing an additional information source in a way of calculating the similarity, so that a certain degree of effect is achieved; however, the additional sources of information are not readily available, especially for users, which involves user privacy, is not only difficult to obtain, but the metadata is protected by law.
Disclosure of Invention
The invention aims to provide an internet content recommendation method and system to solve the problems.
In order to achieve the purpose, the invention provides the following scheme:
an internet content recommendation method comprising:
selecting k from n known users as candidate guides, and initializing the weight of each candidate guide related to the item type; k and n are positive integers greater than 1, and k is less than or equal to n;
training the weight of each candidate leader related to the item type to update the weight of each candidate leader related to the item type, and selecting the candidate leader with the highest weight corresponding to each updated item type as a leader;
and when a new user performs interviewing, recommending the content of the new user through the leader corresponding to the item type selected by the new user.
Preferably, the selecting k from n known users as candidate guides and initializing the weight of each candidate guide related to the item type includes:
establishing a rating training model according to historical interaction data of n known users, wherein the rating training model is as follows:
r ij ′＝u+m i +b j +q j p i T ；
in the formula: r is ij ' represents the predicted rating of the jth item of the ith known user, u represents the average of the true ratings, m i Represents the bias of the ith known user, b j Offset, q, representing the jth item type j Potential factor, p, representing the jth item type i Potential factor, p, representing the ith known user i T Represents p i Transpose of (q) j p i T Represents q j And p i T The inner product between;
updating parameters in the rating training model by adopting a stochastic gradient descent method, wherein the method comprises the following steps:
m i ←m i +α[(r ij -r ij ′)-β*m i ]；
b j ←b j +α[(r ij -r ij ′)-β*b j ]；
q j ←q j +α[(r ij -r ij ′)*p i T -β*q j ]；
p i ←p i +α[(r ij -r ij ′)*q j -β*p i ]；
in the formula: both alpha and beta are hyperparameters, r ij Representing the true rating of the jth item of the ith known user;
obtaining a prediction rating of the item type related to each known user according to the rating training model after parameter updating, obtaining a root mean square error of the item type related to each known user according to the prediction rating and the real rating, and selecting k known users as candidate guides according to the quantity of the item types related to each known user and the root mean square error;
the weight of the item type involved by each candidate leader is initialized.
Preferably, the training the weight of each candidate leader related to the item type to update the weight of each candidate leader related to the item type, and selecting the candidate leader with the highest weight corresponding to each updated item type as the leader includes:
randomly grouping the known users except the candidate leader according to a set proportion to obtain two groups of user sets, and selecting one of the two groups of user sets with a high proportion coefficient as a training user set;
enabling training users in the training user set to randomly select a set number of the item types;
establishing a weight training model as follows:
W yj ←W yj +(a-func(e jy ，e jl ′))；
in the formula: w is a group of yj Weight representing the jth item type of the yth candidate leader, a represents the sum of the root mean square errors of the known users, e jy True rating, e, representing the jth item type of the yth candidate leader jl ' indicates the predicted rating, func (e) for the jth item type of the training user jy ，e jl ′)＝∑(e jy -e jl ′)/|S y |，S y Set of ratings representing the yth candidate leader relating to item types, | S y | denotes S y The number of elements of (a);
obtaining the updated weight of each candidate leader related to the item type according to the weight training model;
and selecting the candidate leader with the highest weight corresponding to each item type as a leader according to the updated weight.
Preferably, when the new user interviews, the content recommendation is performed on the new user through the leader corresponding to the item type selected by the new user, including:
conducting interview on the new user to select the item type;
the leader generates a content recommendation list according to the item type selected by the new user;
and judging the content in the content recommendation list according to set conditions, if the content meets the requirements, recommending the content to the new user, and if the content does not meet the requirements, discarding the content.
Preferably, the setting conditions include:
the real grade of the leader on the item type corresponding to the content in the content recommendation list is higher than a set grade threshold;
and secondly, the rating times of the item types corresponding to the contents in the content recommendation list are larger than the set percentage of the number of the known users.
An internet content recommendation system comprising:
the candidate guide determining and processing module is used for selecting k candidate guides from n known users and initializing the weight of each candidate guide related to the item type; k and n are positive integers greater than 1, and k is less than or equal to n;
the leader determining module is used for training the weight of each candidate leader related to the item type to update the weight of each candidate leader related to the item type, and selecting the candidate leader with the highest weight corresponding to each updated item type as the leader;
and the content recommendation module is used for recommending the content of the new user through the leader corresponding to the item type selected by the new user when the new user interviews.
Preferably, the candidate leader determining and processing module includes:
a rating training model determining unit, configured to establish a rating training model according to historical interaction data of the n known users, as follows:
r ij ′＝u+m i +b j +q j p i T ；
in the formula: r is ij ' represents a predicted rating of the jth item of the ith known user, u represents an average of true ratings, and m i Representing the offset of the ith known user, b j Offset, q, representing the jth item type j Potential factor, p, representing the jth item type i Potential factor, p, representing the ith known user i T Represents p i Transpose of (q) j p i T Denotes q j And p i T The inner product between;
the parameter updating unit is used for updating the parameters in the rating training model by adopting a random gradient descent method, and the following formula is as follows:
m i ←m i +α[(r ij -r ij ′)-β*m i ]；
b j ←b j +α[(r ij -r ij ′)-β*b j ]；
q j ←q j +α[(r ij -r ij ′)*p i T -β*q j ]；
p i ←p i +α[(r ij -r ij ′)*q j -β*p i ]；
in the formula: both alpha and beta are hyperparameters, r ij Representing a true rating of a jth item of an ith known user;
the candidate leader determining unit is used for obtaining a prediction rating of the item type related to each known user according to the rating training model after the parameters are updated, obtaining a root mean square error of the item type related to each known user according to the prediction rating and the real rating, and selecting k known users as candidate leaders according to the quantity of the item type related to each known user and the root mean square error;
and the candidate leader processing unit is used for initializing the weight of the item type related to each candidate leader.
Preferably, the leader determination module includes:
a training user set determining unit, configured to randomly group the known users except the candidate leader according to a set ratio to obtain two user sets, and select one of the two user sets with a high ratio coefficient as a training user set;
the item type selection unit enables training users in the training user set to randomly select the item types with set quantity;
the weight training model determining unit is used for establishing a weight training model as follows:
W yj ←W yj +(a-func(e jy ，e jl ′))；
in the formula: w yj Weight representing the jth item type of the yth candidate leader, a represents the sum of the root mean square errors of the known users, e jy True rating, e, representing the jth item type of the yth candidate leader jl ' indicates the predicted rating, func (e) for the jth item type of the training user jy ，e jl ′)＝∑(e jy -e jl ′)/|S y |，S y Set of ratings representing the yth candidate leader relating to item types, | S y I represents S y The number of elements of (a);
the weight updating unit is used for obtaining the updated weight of each candidate leader related to the item type according to the weight training model;
and the leader determining unit selects the candidate leader with the highest weight corresponding to each item type as a leader according to the updated weight.
Preferably, the content recommendation module includes:
the interview unit is used for interviewing the new user to select the item type;
a content recommendation list determining unit, wherein the leader generates a content recommendation list according to the item type selected by the new user;
and the judging unit is used for judging the contents in the content recommendation list according to set conditions, recommending the contents to the new user if the contents meet the requirements, and discarding the contents if the contents do not meet the requirements.
Preferably, the setting conditions include:
the real grade of the leader on the item type corresponding to the content in the content recommendation list is higher than a set grade threshold;
and secondly, the rating times of the item types corresponding to the contents in the content recommendation list are larger than the set percentage of the number of the known users.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the method comprises the following steps: selecting k from n known users as candidate guides, and initializing the weight of each candidate guide related to the item type; training the weight of each candidate leader related to the item type to update the weight of each candidate leader related to the item type, and selecting the candidate leader with the highest weight corresponding to each updated item type as a leader; and when a new user performs interviewing, recommending the content of the new user through the leader corresponding to the item type selected by the new user. According to the scheme, the problem of cold start of the recommendation system is solved on the basis of not needing metadata of a new user.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
Fig. 1 is a flowchart of an internet content recommendation method according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide an internet content recommendation method and system to solve the problem of cold start of an internet recommendation system.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
As shown in fig. 1, the internet content recommendation method of the present invention includes:
selecting k from n known users as candidate guides, and initializing the weight of each candidate guide related to the type of the item; k and n are positive integers greater than 1, and k is less than or equal to n.
Training the weight of each candidate leader related to the item type to update the weight of each candidate leader related to the item type, and selecting the candidate leader with the highest weight corresponding to each updated item type as the leader.
And when a new user performs interviewing, recommending the content of the new user through the leader corresponding to the item type selected by the new user.
As an alternative embodiment, the selecting k candidate guides from n known users and initializing the weight of each candidate guide related to the item type includes:
establishing a rating training model according to historical interaction data of n known users, wherein the rating training model is as follows:
r ij ′＝u+m i +b j +q j p i T ；
in the formula: r is a radical of hydrogen ij ' represents the predicted rating of the jth item of the ith known user, u represents the average of the true ratings, m i Representing the offset of the ith known user, b j Offset, q, representing the jth item type j Potential factor, p, representing the jth item type i Potential factor, p, representing the ith known user i T Denotes p i Transpose of (q), q j p i T Denotes q j And p i T The inner product between.
Updating parameters in the rating training model by adopting a stochastic gradient descent method, wherein the method comprises the following steps:
m i ←m i +α[(r ij -r ij ′)-β*m i ]；
b j ←b j +α[(r ij -r ij ′)-β*b j ]；
q j ←q j +α[(r ij -r ij ′)*p i T -β*q j ]；
p i ←p i +α[(r ij -r ij ′)*q j -β*p i ]；
in the formula: both alpha and beta are hyperparameters, r ij Representing the true rating of the jth item of the ith known user.
Obtaining a prediction rating of the item type related to each known user according to the rating training model after the parameters are updated, obtaining a root mean square error of the item type related to each known user according to the prediction rating and the real rating, and selecting k known users as candidate guides according to the number of the item types related to each known user and the root mean square error.
Performing descending order arrangement according to the quantity of the known user related item types to obtain a first ordering list; performing ascending arrangement according to the root mean square error to obtain a second ordering list; and selecting the intersection of the first 10% in the first sequence table and the first 10% in the second sequence table as the candidate leader.
The root mean square error is calculated as follows:
in the formula: g ij Root mean square error, S, representing the jth item type of the ith known user i Set of ratings, S, representing the ith known user related to item type i I represents S i The number of elements of (c).
A weight of an item type to which each of the candidate leads relates is initialized.
In this example, the weight average is initialized to 1.
As an optional implementation manner, the training the weight of each candidate leader related to the item type to update the weight of each candidate leader related to the item type, and selecting the candidate leader with the highest weight corresponding to each updated item type as the leader includes:
and randomly grouping the known users except the candidate leader according to a set proportion to obtain two groups of user sets, and selecting one of the two groups of user sets with a high proportion coefficient as a training user set. In this embodiment, the set ratio is 3.
And enabling the training users in the training user set to randomly select the set number of the item types. In this embodiment, the set number is an integer of 2 or more and 6 or less.
Establishing a weight training model as follows:
W yj ←W yj +(a-func(e jy ，e jl ′))；
in the formula: w yj Represents the weight of the jth item type of the jth candidate leader, a representsSum of root mean square errors of known users, e jy A true rating representing the jth item type of the jth candidate leader, e jl ' represents the predicted rating, func (e) for the jth item type of the ith training user jy ，e jl ′)＝∑(e jy -e jl ′)/|S y |，S y Set of ratings, S, representing the y-th candidate leader related to the item type y I represents S y The number of elements of (c).
And obtaining the updated weight of each candidate leader related to the item type according to the weight training model.
And selecting the candidate leader with the highest weight corresponding to each item type as a leader according to the updated weight.
As an optional implementation manner, when the new user performs interviewing, the content recommendation performed on the new user by the leader corresponding to the item type selected by the new user includes:
the new user is interviewed to select a type of item.
And the leader generates a content recommendation list according to the item type selected by the new user.
And judging the contents in the content recommendation list according to set conditions, recommending to the new user if the contents meet the requirements, and discarding if the contents do not meet the requirements.
As an optional implementation manner, the setting conditions of the present invention include:
one is that the real score of the leader on the item type corresponding to the content in the content recommendation list is higher than a set score threshold.
And secondly, the rating times of the item types corresponding to the contents in the content recommendation list are larger than the set percentage of the number of the known users. In this embodiment, the set percentage is 5%.
The invention also provides an internet content recommendation system, comprising:
the candidate leader determining and processing module is used for selecting k candidate leaders from n known users and initializing the weight of each candidate leader related to the item type; k and n are positive integers greater than 1, and k is less than or equal to n.
And the leader determining module is used for training the weight of each candidate leader related to the item type so as to update the weight of each candidate leader related to the item type, and selecting the candidate leader with the highest weight corresponding to each updated item type as the leader.
And the content recommendation module is used for recommending the content to the new user through the guide corresponding to the item type selected by the new user when the new user interviews.
As an optional implementation manner, the candidate leader determining and processing module of the present invention includes:
a rating training model determining unit, configured to establish a rating training model according to historical interaction data of the n known users, as follows:
r ij ′＝u+m i +b j +q j p i T ；
in the formula: r is a radical of hydrogen ij ' represents a predicted rating of the jth item of the ith known user, u represents an average of true ratings, and m i Representing the offset of the ith known user, b j Offset, q, representing the jth item type j Potential factor, p, representing the jth item type i Potential factor, p, representing the ith known user i T Represents p i Transpose of (q) j p i T Denotes q j And p i T The inner product between.
The parameter updating unit is used for updating the parameters in the rating training model by adopting a random gradient descent method, and the following formula is as follows:
m i ←m i +α[(r ij -r ij ′)-β*m i ]；
b j ←b j +α[(r ij -r ij ′)-β*b j ]；
q j ←q j +α[(r ij -r ij ′)*p i T -β*q j ]；
p i ←p i +α[(r ij -r ij ′)*q j -β*p i ]；
in the formula: both alpha and beta are hyperparameters, r ij Representing the true rating of the jth item of the ith known user.
And the candidate leader determining unit is used for obtaining a prediction rating of the item type related to each known user according to the rating training model after the parameters are updated, obtaining a root mean square error of the item type related to each known user according to the prediction rating and the real rating, and selecting k known users as candidate leaders according to the quantity of the item types related to each known user and the root mean square error.
And the candidate leader processing unit is used for initializing the weight of the item type related to each candidate leader.
As an optional embodiment, the leader determining module of the present invention includes:
and the training user set determining unit is used for randomly grouping the known users except the candidate leader according to a set proportion to obtain two groups of user sets, and selecting one group with a high proportion coefficient from the two groups of user sets as a training user set.
And the item type selection unit enables the training users in the training user set to randomly select the set number of item types.
The weight training model determining unit is used for establishing a weight training model as follows:
W yj ←W yj +(a-func(e jy ，e jl ′))；
in the formula: w yj Weight representing the jth item type of the yth candidate leader, a represents the sum of the root mean square errors of the known users, e jy True rating, e, representing the jth item type of the yth candidate leader jl ' represents the predicted rating, func (e) for the jth item type of the ith training user jy ，e jl ′)＝∑(e jy -e jl ′)/|S y |，S y Set of ratings representing the yth candidate leader relating to item types, | S y I represents S y The number of elements of (c).
And the weight updating unit is used for obtaining the updated weight of each candidate leader related to the item type according to the weight training model.
And the leader determining unit selects the candidate leader with the highest weight corresponding to each item type as a leader according to the updated weight.
As an optional implementation manner, the content recommendation module of the present invention includes:
and the interview unit is used for interviewing the new user to select the item type.
And the guide generates a content recommendation list according to the item type selected by the new user.
And the judging unit is used for judging the contents in the content recommendation list according to set conditions, recommending the contents to the new user if the contents meet the requirements, and discarding the contents if the contents do not meet the requirements.
The method and the system provided by the invention have the advantages that the content recommendation is carried out on the new user by the leader, the cold start problem of the internet content recommendation system is solved, and the metadata of the user is not needed.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
The principle and the embodiment of the present invention are explained by applying specific examples, and the above description of the embodiments is only used to help understanding the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (6)
1. An internet content recommendation method, comprising:
selecting k from n known users as candidate guides, and initializing the weight of each candidate guide related to the type of the item; k and n are positive integers greater than 1, and k is less than or equal to n;
training the weight of each candidate leader related to the item type to update the weight of each candidate leader related to the item type, and selecting the candidate leader with the highest weight corresponding to each updated item type as a leader;
when a new user conducts interview, content recommendation is conducted on the new user through the guide corresponding to the item type selected by the new user;
selecting k from n known users as candidate guides, and initializing the weight of each candidate guide related to the item type, wherein the weight comprises the following steps:
establishing a rating training model according to historical interaction data of n known users, wherein the rating training model is as follows:
r′ ij ＝u+m i +b j +q j p i T ；
in the formula: r is ij ' represents the predicted rating of the jth item of the ith known user, u represents the average of the true ratings, m i Representing the offset of the ith known user, b j Offset, q, representing the jth item type j Potential factor, p, representing the jth item type i Potential factor, p, representing the ith known user i T Denotes p i Transpose of (q) j p i T Denotes q j And p i T The inner product between;
updating parameters in the rating training model by adopting a stochastic gradient descent method, wherein the method comprises the following steps:
m i ←m i +α[(r ij -r′ ij )-β*m i ]；
b j ←b j +α[(r ij -r′ ij )-β*b j ]；
q j ←q j +α[(r ij -r′ ij )*p i T -β*q j ]；
p i ←p i +α[(r ij -r′ ij )*q j -β*p i ]；
in the formula: both alpha and beta are hyperparameters, r ij Representing a true rating of a jth item of an ith known user;
obtaining a prediction rating of the item type related to each known user according to the rating training model after the parameters are updated, obtaining a root mean square error of the item type related to each known user according to the prediction rating and the real rating, and selecting k known users as candidate guides according to the number of the item types related to each known user and the root mean square error;
initializing the weight of the item type related to each candidate leader;
the training of the weight of each candidate leader related to the item type to update the weight of each candidate leader related to the item type, and selecting the candidate leader with the highest weight corresponding to each updated item type as the leader, includes:
randomly grouping the known users except the candidate leader according to a set proportion to obtain two groups of user sets, and selecting one of the two groups of user sets with a high proportion coefficient as a training user set;
enabling training users in the training user set to randomly select a set number of the item types;
establishing a weight training model as follows:
W yj ←W yj +(a-func(e jy ，e jl ′))；
in the formula: w yj Weight representing the jth item type of the yth candidate leader, a represents the sum of the root mean square errors of the known users, e jy True rating, e, representing the jth item type of the yth candidate leader jl ' represents the predicted rating, func (e) for the jth item type of the ith training user jy ，e jl ′)＝∑(e jy -e jl ′)/|S y |，S y Set of ratings, S, representing the y-th candidate leader related to the item type y I represents S y The number of elements of (a);
obtaining the updated weight of each candidate leader related to the item type according to the weight training model;
and selecting the candidate leader with the highest weight corresponding to each item type as a leader through the updated weight.
2. The method as claimed in claim 1, wherein when the new user performs interview, the content recommendation is performed on the new user by the leader corresponding to the item type selected by the new user, and the method comprises:
performing interview on the new user to select a project type;
the leader generates a content recommendation list according to the item type selected by the new user;
and judging the content in the content recommendation list according to set conditions, if the content meets the requirements, recommending the content to the new user, and if the content does not meet the requirements, discarding the content.
3. The method as claimed in claim 2, wherein the setting condition comprises:
the real grade of the leader on the item type corresponding to the content in the content recommendation list is higher than a set grade threshold;
and secondly, the rating times of the item types corresponding to the contents in the content recommendation list are larger than the set percentage of the number of the known users.
4. An internet content recommendation system, comprising:
the candidate guide determining and processing module is used for selecting k candidate guides from n known users and initializing the weight of each candidate guide related to the item type; k and n are positive integers greater than 1, and k is less than or equal to n;
the leader determining module is used for training the weight of each candidate leader related to the item type to update the weight of each candidate leader related to the item type, and selecting the candidate leader with the highest weight corresponding to each updated item type as the leader;
the content recommendation module is used for recommending the content to the new user through the leader corresponding to the item type selected by the new user when the new user interviews;
the candidate leader determination and processing module comprising:
a rating training model determining unit, configured to establish a rating training model according to historical interaction data of the n known users, as follows:
r ij ′＝u+m i +b j +q j p i T ；
in the formula: r is ij ' represents a predicted rating of the jth item of the ith known user, u represents an average of true ratings, and m i Represents the bias of the ith known user, b j Offset, q, representing the jth item type j Potential factor, p, representing the jth item type i Potential factor, p, representing the ith known user i T Denotes p i Transpose of (q) j p i T Represents q j And p i T The inner product between;
the parameter updating unit is used for updating the parameters in the rating training model by adopting a random gradient descent method, and the following formula is as follows:
m i ←m i +α[(r ij -r ij ′)-β*m i ]；
b j ←b j +α[(r ij -r ij ′)-β*b j ]；
q j ←q j +α[(r ij -r ij ′)*p i T -β*q j ]；
p i ←p i +α[(r ij -r ij ′)*q j -β*p i ]；
in the formula: both alpha and beta are hyperparameters, r ij Representing a true rating of a jth item of an ith known user;
the candidate leader determining unit is used for obtaining a prediction rating of the item type related to each known user according to the rating training model after the parameters are updated, obtaining a root mean square error of the item type related to each known user according to the prediction rating and the real rating, and selecting k known users as candidate leaders according to the quantity of the item type related to each known user and the root mean square error;
a candidate leader processing unit for initializing a weight of an item type to which each of the candidate leaders relates;
the leader determination module, comprising:
a training user set determining unit, configured to randomly group the known users except the candidate leader according to a set ratio to obtain two user sets, and select one of the two user sets with a high proportionality coefficient as a training user set;
the item type selection unit enables training users in the training user set to randomly select the item types with set quantity;
the weight training model determining unit is used for establishing a weight training model as follows:
in the formula: w yj Weight representing the jth item type of the yth candidate leader, a represents the sum of the root mean square errors of the known users, e jy A true rating representing the jth item type of the jth candidate leader, e jl ' represents the predicted rating, func (e) for the jth item type of the ith training user jy ，e jl ′)＝∑(e jy -e jl ′)/|S y |，S y Set of ratings, S, representing the y-th candidate leader related to the item type y I represents S y Element (1)The number of elements;
the weight updating unit is used for obtaining the updated weight of each candidate leader related to the item type according to the weight training model;
and the leader determining unit selects the candidate leader with the highest weight corresponding to each item type as a leader according to the updated weight.
5. The internet content recommendation system according to claim 4, wherein said content recommendation module comprises:
the interview unit is used for interviewing the new user to select the item type;
a content recommendation list determining unit, wherein the leader generates a content recommendation list according to the item type selected by the new user;
and the judging unit is used for judging the contents in the content recommendation list according to set conditions, recommending the contents to the new user if the contents meet the requirements, and discarding the contents if the contents do not meet the requirements.
6. The internet content recommendation system according to claim 5, wherein the setting condition comprises:
the real grade of the leader on the item type corresponding to the content in the content recommendation list is higher than a set grade threshold;
and secondly, the rating times of the item types corresponding to the contents in the content recommendation list are larger than the set percentage of the number of the known users.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010089194.2A CN111259266B (en) | 2020-02-12 | 2020-02-12 | Internet content recommendation method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010089194.2A CN111259266B (en) | 2020-02-12 | 2020-02-12 | Internet content recommendation method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111259266A CN111259266A (en) | 2020-06-09 |
| CN111259266B true CN111259266B (en) | 2023-04-07 |
Family
ID=70945605
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010089194.2A Active CN111259266B (en) | 2020-02-12 | 2020-02-12 | Internet content recommendation method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111259266B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105893569A (en) * | 2016-03-31 | 2016-08-24 | 北京奇虎科技有限公司 | News weight calculation method and apparatus |
| CN109635291A (en) * | 2018-12-04 | 2019-04-16 | 重庆理工大学 | A kind of recommended method of fusion score information and item contents based on coorinated training |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9760609B2 (en) * | 2013-11-22 | 2017-09-12 | Here Global B.V. | Graph-based recommendations service systems and methods |
-
2020
- 2020-02-12 CN CN202010089194.2A patent/CN111259266B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105893569A (en) * | 2016-03-31 | 2016-08-24 | 北京奇虎科技有限公司 | News weight calculation method and apparatus |
| CN109635291A (en) * | 2018-12-04 | 2019-04-16 | 重庆理工大学 | A kind of recommended method of fusion score information and item contents based on coorinated training |
Non-Patent Citations (3)
| Title |
|---|
| 3GPP. "Virtual Reality (VR) media services over 3GPP".《3GPP TR 26.918 V15.0.0》.2017, * |
| 人工智能时代新闻内容分发中的媒介伦理问题研究;何慧敏;《新媒体研究》(第21期);全文 * |
| 虚实融合场景中多光源的位置与强度估计研究;徐超;《CNKI优秀硕士学位论文全文库》;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111259266A (en) | 2020-06-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sun et al. | Social-aware video recommendation for online social groups | |
| US11151209B1 (en) | Recommending objects to a user of a social networking system based on the location of the user | |
| WO2018090793A1 (en) | Multimedia recommendation method and device | |
| US10747771B2 (en) | Method and apparatus for determining hot event | |
| CN104199896B (en) | The video similarity of feature based classification is determined and video recommendation method | |
| US9734210B2 (en) | Personalized search based on searcher interest | |
| Zhao et al. | Service quality evaluation by exploring social users’ contextual information | |
| Liang et al. | Time-aware topic recommendation based on micro-blogs | |
| US20190278819A1 (en) | Systems and methods of providing recommendations of content items | |
| US20190340176A1 (en) | System and method for data mining and similarity estimation | |
| JP5738785B2 (en) | Recommendation program, apparatus and method for recommending critics that meet user preferences | |
| Dhawan et al. | High rating recent preferences based recommendation system | |
| KR101725510B1 (en) | Method and apparatus for recommendation of social event based on users preference | |
| KR101708254B1 (en) | Story-based recommendation system and method for movies by character-net and collaborative filtering | |
| Liu et al. | Personalised context-aware re-ranking in recommender system | |
| US20140289334A1 (en) | System and method for recommending multimedia information | |
| US8984091B1 (en) | Providing content based on timestamp of last request for content | |
| US8745074B1 (en) | Method and system for evaluating content via a computer network | |
| CN111259266B (en) | Internet content recommendation method and system | |
| Sacco et al. | In users we trust: towards social user interactions based Trust Assertions for the Social Semantic Web | |
| Zhang et al. | Estimating online review helpfulness with probabilistic distribution and confidence | |
| CN111666481B (en) | Data mining method, device, computer readable medium and electronic equipment | |
| Sood et al. | Preference based personalized news recommender system | |
| Chao et al. | A Social Network system oriented hybrid recommendation model | |
| Reafee et al. | The social network role in improving recommendation performance of collaborative filtering |
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 |