CN108319622A - A kind of media content recommendations method and device - Google Patents
A kind of media content recommendations method and device Download PDFInfo
- Publication number
- CN108319622A CN108319622A CN201710037620.6A CN201710037620A CN108319622A CN 108319622 A CN108319622 A CN 108319622A CN 201710037620 A CN201710037620 A CN 201710037620A CN 108319622 A CN108319622 A CN 108319622A
- Authority
- CN
- China
- Prior art keywords
- media content
- similarity
- user
- alternative
- media
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Present applicant proposes a kind of media content recommendations methods, on the basis of obtaining alternative media content by collaborative filtering, user's similitude between each alternative media content and the media content of user's browsing, content similarities are further considered to obtaining associated recommendation media content after the progress of alternative media content further filtering, excavate more correlations based on media content and readable media content.The application also proposed corresponding media content recommendations device.
Description
Technical field
This application involves Internet technical field more particularly to a kind of media content recommendations method and devices.
Background technology
Currently, with the fast development of internet, web database technology constantly increases, and is brought obtaining information to the network user
Problem of information overload is also resulted in while convenient, how fast and effeciently to be searched and located to the letter needed in the data of magnanimity
Breath becomes the developing outstanding problem of current internet, and the hot spot of networked information retrieval research.
To solve the above-mentioned problems, many media content platforms user access browse a media content in other words when, to
User recommends relevant other media contents.Such as:News website all can be by new after user opens a certain news content
Other relevant news of news that the way of recommendation recommends currently to show with news client to user are heard, are read as extension.Newly
It can be that user recommends Personalize News information to hear the mode recommended, and help user to find interested content, can effectively help
It helps user quickly and accurately to find the resource of needs, has broad application prospects.
Invention content
This application provides a kind of media content recommendations methods, including:It obtains in predetermined amount of time for each media content
User access record, and record is accessed according to the user and determines the first similarity between each two media content and deposits
Storage, the similarity of the access user of first similarity characterization, two media contents;Receive the media that applications client is sent
Content recommendation request, wherein when the first media content in each media content is visited in the applications client by user
When asking, the applications client sends the media content recommendations request;It is asked in response to the media content recommendations, from being deposited
The first similarity of each second media content and first media content is obtained in first similarity of storage, will with it is described
First similarity of the first media content is more than each second media content of the first predetermined threshold value as alternative media content;It calculates
The second similarity between each alternative media content and first media content, second similarity characterization, two media
The content similarity of content;Based in the alternative media content between each alternative media content and first media content
The first similarity and each alternative media content of the second similarity calculation recommendability score, and recommendability score is surpassed
Cross associated recommendation media content of the alternative media content of the second predetermined threshold value as first media content;By described first
The link of the associated recommendation media content of media content is sent to the applications client.
The application also proposed a kind of media content recommendations device, including:First similarity determining unit, it is pre- for obtaining
Record is accessed for the user of each media content in section of fixing time, and record is accessed according to the user and is determined in each two media
The first similarity between appearance simultaneously stores, the similarity of the access user of first similarity characterization, two media contents;Matchmaker
Body content recommendation request receiving unit, the media content recommendations request for receiving applications client transmission, wherein when described each
When the first media content in media content is accessed by the user in the applications client, described in the applications client transmission
Media content recommendations are asked;Alternative media content determining unit, for being asked in response to the media content recommendations, from being stored
First similarity in obtain the first similarity of each second media content and first media content, will be with described
First similarity of one media content is more than each second media content of the first predetermined threshold value as alternative media content;Second phase
It is described for calculating the second similarity between each alternative media content and first media content like degree computing unit
The content similarity of second two media contents of similarity characterization;Associated recommendation media content determination unit, for based on described
The first similarity and the second similarity in alternative media content between each alternative media content and first media content
The recommendability score of each alternative media content is calculated, and is more than the alternative media of the second predetermined threshold value by recommendability score
Associated recommendation media content of the content as first media content;Transmission unit is used for first media content
The link of associated recommendation media content is sent to the applications client.
The said program proposed using the application, can obtain the stronger correlation of correlation can recommend media content.
Description of the drawings
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention without having to pay creative labor, may be used also for those of ordinary skill in the art
With obtain other attached drawings according to these attached drawings.
Fig. 1 is the system architecture schematic diagram that the media content recommendations method that present application example proposes is related to;
Fig. 2 be this application involves a kind of user interface schematic diagram;
Fig. 3 is the flow diagram for the media content recommendations method that present application example proposes;
Fig. 4 is Collaborative Filtering Recommendation Algorithm and contrast schematic diagram of the content-based recommendation algorithm about clicking rate;
Fig. 5 is to calculate the flow diagram for obtaining the first similarity;
Fig. 6 is using recommendation results coverage rate schematic diagram after size windows parallel computation;
Fig. 7 is the flow diagram that the first similarity is calculated using cosine similarity;
Fig. 8 is the flow diagram that the second similarity is calculated using cosine similarity;
Fig. 9 is the schematic diagram using clicking rate after duplicate removal strategy;
Figure 10 is the structural schematic diagram for the media content recommendations device that present application example proposes;And
Figure 11 is the composite structural diagram of the computing device where the media content recommendations device that present application example proposes.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation describes, it is clear that described embodiment is only a part of the embodiment of the present invention, instead of all the embodiments.Based on this
Embodiment in invention, every other reality obtained by those of ordinary skill in the art without making creative efforts
Example is applied, shall fall within the protection scope of the present invention.
Present applicant proposes a kind of media content recommendations methods Internet-based, and this method can be applied to shown in FIG. 1
In system architecture.As shown in Figure 1, the system architecture includes:Using (APP) client 101, pushed information platform 102 and push
Information providing client 105, these entities can be communicated by internet 106, and wherein pushed information platform 102 includes
Application server 103 and user access database of record 104.
Terminal user can use applications client 101 to access the application server 103 in pushed information platform 102, than
Such as:Browse news or article etc..When user accesses application server 103 using applications client 101, applications client
101 can be reported to the access behavior of user application server 103, and application server 103 is by the access behavioral data of user
User is stored in access in database of record 104.While 101 report of user of applications client accesses behavior, applications client
101 can send out information push request to pushed information platform 102, and pushed information platform 102 can will be pushed with the information and be asked
The media content to match is asked to be pushed to applications client 101.By pushed information provider client 105, pushed information carries
The material for the media content that it to be pushed can be uploaded to pushed information platform 102 by supplier, to generate accordingly for pushing
Media content.
When above-mentioned media content is news, system architecture shown in FIG. 1 can be the system architecture realized news and recommended,
Wherein, pushed information platform 102 can be news push platform, and pushed information provider can be news briefing person, using visitor
Family end 101 is news client, and application server 103 is NEWS SERVER.Fig. 2 shows be a news APP client
The page, the news browsed which show a user and lower section it is new with relevant 3 recommendations of the news of the browsing
Message is heard, the news messages that each is recommended include title and picture, click the title and picture of every recommendation news messages, answer
The complete content of the recommendation news messages is shown with client 101.When user uses news Client browse news, news visitor
Family end sends out news push request to news push platform, news push platform by with the news that is just browsed it is relevant recommend it is new
The link of news is sent to news client, which is illustrated in the form of word or picture under browsing news by news client
The related of side is read, when the user clicks the word or when picture, and the displaying of news client recommends the full content of news.
In some instances, pushed information platform 102 is pushing away based on content when obtaining associated recommendation media content
It recommends, according to the keyword message for the media content that user is browsing, keyword message relevance scores is met into threshold condition
Media content is as associated recommendation media content.Current content-based recommendation has the following defects:Simple dependence content
Correlation, the media content quality for recommending out are difficult to ensure, while being susceptible to old text in the recommendation of the article of such as political situation of the time class
The case where Deng user experience is influenced.
Based on above-mentioned technical problem, the application proposes that a kind of media content recommendations method, this method can be applied to push letter
Platform 102 is ceased, as shown in figure 3, this approach includes the following steps:
Step 301:It obtains the user in predetermined amount of time for each media content and accesses record, and visited according to the user
It asks the first similarity between record determines each two media content and stores, first similarity characterization, two media contents
Access user similarity.
In this step, the user for obtaining each media content first accesses record, obtains in nearest predetermined amount of time and owns
User records the access of each media content.Terminal user using applications client 101 when accessing application server, using visitor
Family end 101 is to the access behavioral data of 103 report of user of application server, and application server 103 is by the access behavior number of user
It is accessed in database of record 104 according to user is stored in.In this step, application server 103 accesses database of record from user
All users record the access of each media content in middle acquisition predetermined amount of time.The behavior record data format of each user
Can be (user A, media content 1, media content 3, media content 5 ...), wherein user A can be the user User ID,
Such as may include that user A registers the various accounts used on various APP, website, such as:The instant messagings such as QQ number, e-
The addresses mail, wechat account, microblog account, Taobao's account etc..
After the user's access record for obtaining each media content in some instances, dirty data is filtered, such as in the user of acquisition
Access the access record that apparent non-user behavior is removed in record and/or the access record for influencing smaller user.Apparent non-use
The access record of family behavior refers to access broad medium content within a predetermined period of time, exceeds the access ability range of normal users,
The access behavior record of some machines.The access record for influencing smaller user refers to access media content amount spy in predetermined amount of time
The access of not small user records, and the user that a media content is such as only accessed in 12 hours accesses record.It is dirty effectively filtering
The first-level class based on media content calculates the first similarity of each two media content on the basis of data, is not herein
The first similarity between any two media content is calculated, but first is calculated between the media content under identical first-level class
Similarity, because with the correlation with bigger between the media content under identical first-level class, under different first-level class
Correlation between media content is smaller.Such as news media's content, the first-level class of news media's content can be with
All there is larger correlation for two news of sport including sport, News & Activitics, amusement, finance and economics etc., such as first-level class,
Thus first order calculation is classified as the first similarity between the news two-by-two under sport.But the news and one that first-level class is sport
The general correlation of news that grade is classified as News & Activitics is smaller, thus does not calculate the first phase between the news under different first-level class
Like degree, which also improves the efficiency of calculating.
Step 302:Receive the media content recommendations request that applications client is sent, wherein when in each media content
The first media content when being accessed by the user in the applications client, the applications client sends the media content and pushes away
Recommend request.
When user accesses the first media content using applications client 101, applications client 101 is to pushed information platform
102 send the message of media content recommendations request.
Step 303:It is asked in response to the media content recommendations, each the is obtained from first similarity stored
First similarity of two media contents and first media content will be more than with the first similarity of first media content
Each second media content of first predetermined threshold value is as alternative media content.
After pushed information platform 102 receives the media content recommendations request message of the transmission of applications client 101, from step
Obtained in 301 and obtained in the first similarity between each two media content that stores each second media content with it is described
First similarity of the first media content, and using first similarity be more than the first predetermined threshold value each second media content as
Alternative media content, to complete the first step filtering of associated recommendation media content.Obtain the mode of the alternative media content
For project-based collaborative filtering mode, the similitude between article is calculated according to the history access record of all users, it will be with
The similar article of article that user likes recommends user.Fig. 4 shows the recommendation of the project-based collaborative filtering of use and shows
Difference of some content-based recommendations in clicking rate, wherein the recommendation of project-based collaborative filtering and pushing away based on content
Recommend each dispensing 50%, it can be seen that compared with content-based recommendation, using the click of the recommendation of project-based collaborative filtering
Rate improves a lot.It can also be obtained in alternative media by the way of based on commending contents in other examples of the application
Hold, the collaborative filtering mode for being also based on user obtains alternative media content or content-based recommendation mode, is based on
The collaborative filtering mode of user, the mode of project-based collaborative filtering mode three fitting obtain alternative media content.When obtaining
When the quantity of the alternative media content obtained is larger, 100 before being taken according to its first similarity in obtaining alternative media content
.
Step 304:Calculate the second similarity between each alternative media content and first media content, described
The content similarity of two two media contents of similarity characterization.
Each media content has some keywords, these keywords are there are in the title of media content or content, root
The media content can be retrieved according to the keyword of media content, the same keyword of two media contents is more, then its content
Similarity is bigger.
Step 305:Based in the alternative media content between each alternative media content and first media content
The first similarity and each alternative media content of the second similarity calculation recommendability score, and recommendability score is surpassed
Cross associated recommendation media content of the alternative media content of the second predetermined threshold value as first media content.
When carrying out related news recommendation, the application be different from it is traditional recommended based on the similitude of content, the application
The media content recommendations method of proposition considers the first similarity and the second similarity of alternative media content, additionally can be with
Consider the temperature of each alternative media content and stylish degree.
Step 306:The link of the associated recommendation media content of first media content is sent to applications client.
After pushed information platform obtains the associated recommendation media content for the media content that user is browsing, by associated recommendation
The link of media content is sent to applications client, which is illustrated in user by applications client in the form of word or picture
During related below the media content just browsed is read, the word or when picture when the user clicks, applications client shows phase
Close the full content for recommending media content.
Using media content recommendations method provided by the present application, on the basis for obtaining alternative media content by collaborative filtering
On, user's similitude, the content further considered between each alternative media content and the media content that user browses is similar
Property to alternative media content carry out further filter after obtain associated recommendation media content, excavate based on media content more
More correlations and readable stronger media content.
In some instances, in above-mentioned steps 301, the determining each two media content of record is accessed according to user executing
Between the first similarity when, as shown in figure 5, can further comprise the steps:
Step 501:Record is accessed according to the user of each media content in nearest first preset time period, calculates each two matchmaker
The first similarity between holding in vivo often undergoes a period 1 and recalculates once, and first similarity is stored
In the first aggregate, the period 1 is less than first preset time.
In this step, first preset time is preset first, according to the user in the first nearest preset time period
Record is accessed to calculate the first similarity between two media contents, while often undergoing a period 1 and recalculating one
It is secondary.First preset time can be 12 hours, and the period 1 can be 1 hour, accessed and recorded according to nearest 12 hours users
The first similarity between two media contents is calculated, is recalculated per hour once, and the first similarity that will be calculated
Storage is in the first aggregate.It calculates the first similarity in this step to calculate by big window, the first preset time is longer, for certain
It is more comprehensive that other media contents similar with the media content are calculated in one media content.But during big window calculates
Period 1 is also long, and the longer update cycle is difficult to meet the related reading need of the explosive media content of fast propagation
It asks.Such as in the case of the update cycle is 1 small, a shocking news was gone out within 1 hour period, when user browses
When the shocking news, with the relevant recommendation news of the shocking news there are no calculating in big window, thus with regard to nothing
Method carries out the shocking news recommendation of related news.To solve the problems, such as this, the application uses while big window calculates
Wicket calculates, specifically as described in step 502.
Step 502:Record is accessed according to the user of each media content in nearest second preset time period, calculates each two matchmaker
The first similarity between holding in vivo often undergoes a second round and recalculates once, and first similarity is stored
In second set, the second round is less than second preset time, and it is pre- that second preset time is less than described first
If the time, the second round is less than the period 1.
The first similarity that the step calculates between two media contents is calculated by wicket, and the second of wicket is default
Time is less than the first preset time of big window, while the second round of wicket is less than the period 1 of big window.Such as second
Preset time is 1 hour, and second round is 10 minutes, and accessing record according to nearest 1 hour user in wicket calculates two
The first similarity between media content, update in every 10 minutes calculates once, and the first similarity being calculated is stored in
In second set.To which when big window is also in calculating, synchronous wicket data, reach meter caused by making up big window in time
Calculate the purpose of delay.
After obtaining the first similarity between every two media content, when the first media content in applications client by user
When access, applications client sends media content recommendations and asks to give pushed information platform.It is described obtain each second media content with
First similarity of first media content includes:It is each that pushed information platform searches acquisition in first set and second set
First similarity of the second media content and first media content, described in face step specific as follows.
Step 503:Judge each second media content and first media content whether are stored in the first set
The first similarity.
Judge the first phase that each second media content and first media content whether are stored in the first set
Like degree, if so, searching the first phase for obtaining each second media content and first media content in the first set
Like degree;Otherwise it is similar to the first of first media content that each second media content of acquisition is searched in the second set
Degree.When determining first similarity of each second media content with first media content, first searches, see in the first aggregate
Whether big window has calculated the first similarity of each second media content and first media content.If first set
It is middle there are the first similarity of each second media content and first media content, then follow the steps 504, in the first aggregate
The first similarity for obtaining each second media content and first media content is searched, it is no to then follow the steps 505, in the second collection
The first similarity for obtaining each second media content and first media content is searched in conjunction
Step 504:It is searched in the first set and obtains the of each second media content and first media content
One similarity.The first similarity-rough set that big window is calculated is comprehensive, thus has calculated the first similarity in big window
In the case of preferentially in the corresponding first set of big window search obtain the first similarity.
Step 505:It is searched in the second set and obtains the of each second media content and first media content
One similarity.In the case where big window calculates the first similarity not yet, searched in the corresponding second set of wicket
Obtain the first similarity.
The case where the first similarity-rough set that big window is calculated is comprehensive, and there are computation delays, although wicket is counted
The first obtained similarity is not comprehensive enough, but can timely update, when project-based collaborative filtering uses big-wicket simultaneously
After row calculates, as shown in fig. 6, the associated recommendation result coverage rate of media content improves.
In some instances, in above-mentioned steps 501 and step 502, between executing calculating each two media content
When the first similarity, the first similarity is calculated using cosine similarity, as shown in fig. 7, can further comprise the steps:
Step 701:Obtain the access user vector of each media content in two media contents.
Record is accessed according to the user in the predetermined amount of time obtained in the above, it is assumed that in the predetermined amount of time
It shares 10 users and has accessed media content, respectively user A, user B, user C, user D, user E, user F, user G, use
Family H, user I, user J.For media content i and media content j, have accessed media content i's in the predetermined amount of time
User includes user A, user B, user D, user F, user G, user I, and media content j is had accessed in the predetermined amount of time
User include user A, user B, user C, user D, user I, user J.User vector characterization is accessed in a media
Which user is appearance have accessed it in the predetermined amount of time, thus access user vector corresponding with media content i
For (1,1,0,1,0,1,1,0,1,0), access user vector corresponding with media content j be (1,1,1,1,0,0,0,0,1,
1)。
Step 702:Calculate the cosine similarity of the access user vector of two media contents.
The access user vector of media content i and media content j as described above, media content i be (1,1,0,1,0,
1,1,0,1,0), the access user vector of media content j is (1,1,1,1,0,0,0,0,1,1), the access user of media content i
The cosine similarity of vector and media content j accessed between user vector is indicated with following formula (1):
The cosine similarity between media content i and media content j can be calculated by formula (1).
Step 703:Using the cosine similarity being calculated as the first similarity between described two media contents.
In some instances, in above-mentioned steps 304, each alternative media content and first media are calculated executing
When the second similarity between content, the second similarity is calculated using cosine similarity, as shown in figure 8, can further comprise
Following steps:
Step 801:Obtain the keyword vector of each alternative media content.
Correspondence between the keyword for all media contents that all users check is as follows:
Media content:Tag1, tag2, tag3 ..., tagM
Media content indicates that all media contents that all users check, tag expressions are carried from each media content in above formula
The keyword of taking-up, keyword are included in media content title or content, can be retrieved in the media according to the keyword
Hold, keyword can be any Chinese, English, number, or the mixture of Chinese English digital.Tag1 indicates that all users check
All news in first keyword, tag2 indicates second keyword in all news that all users check,
Tag3 indicates the third keyword in all news that all users check, and so on, tagM indicates that all users check
All media contents in m-th keyword, M indicates the number of the whole keywords for all media contents that all users check
Amount.All include which keyword according to a media content, it may be determined that keyword vector corresponding with the media content,
Such as media content i includes tag1 and tag3, then keyword vector corresponding with media content i be (1,0,1,0,0,
0 ...).
Step 802:Obtain the keyword vector of the first media content.Obtain the side of the keyword vector of the first media content
Formula is identical as each mode of keyword vector of alternative media content of above-mentioned acquisition, and details are not described herein.
Step 803:Calculate the keyword of the keyword vector of first media content and each alternative media content to
The cosine similarity of amount.
When the keyword vector of an alternative media content isThe keyword vector of first media content isWhen, it is crucial
Word vectorAnd keyword vectorCosine similarity pass through formula (1) calculate.
Step 804:Using the cosine similarity being calculated as each alternative media content and first media
The second similarity between content.
In some instances, it in above-mentioned steps 305, is executing based on each alternative media in the alternative media content
The first similarity and each alternative media content of the second similarity calculation between content and first media content push away
When the property recommended score, the first similarity of each alternative media content and the second Similarity-Weighted can be summed to obtain each candidate
The recommendability score of media content.Wherein the weight of the weight of the first similarity and the second similarity can in advance be set by experience
It is fixed, it can also be obtained by machine learning, and be adjusted according to sampling results.
In other examples, in above-mentioned steps 305, executing based on each candidate matchmaker in the alternative media content
In vivo hold and first media content between the first similarity and each alternative media content of the second similarity calculation can
When recommendatory score, it may comprise steps of:
1) temperature of each alternative media content and stylish degree, are obtained.Some media contents itself have hot spot attribute, right
Media content adds hot value in this way.Stylish degree is to decay for the time of present system time the time that media content generates
Degree.
2), by the first similarity of each alternative media content, the second similarity, temperature and/or stylish degree weighted sum
Obtain the recommendability score of each alternative media content.
When the factor of consideration include the first similarity, the second similarity, temperature and it is stylish spend when, using following formula (2)
Calculate the recommendability score of each alternative media content.
Score=a1WCF+a2WTag+a3WHot+a4WTime(ai> 0) (2)
In formula (2), WCFFor the first similarity, WTagFor the second similarity, WHotFor temperature, WTimeFor stylish degree, α1、
α2、α3、α4, be respectively the first similarity, the weight parameter of the second similarity, temperature and stylish degree.Each candidate matchmaker is calculated
It is more than the alternative media content of the second predetermined threshold value as described using recommendability score after the recommendability score held in vivo
The associated recommendation media content of one media content.
In some instances, the media content recommendations method of the application proposition further includes:
The associated recommendation media content being accessed by the user is rejected in the associated recommendation media content.
The principle of associated recommendation often causes that feature is similar or stronger two media contents of relevance calculate
Associated recommendation media content registration is very high, and the related of different media contents is caused to read the identical associated recommendation matchmaker of multiple exposure
Hold in vivo, influences the exposing of user experience and long-tail article.The exposure of extra storage associated recommendation media content is needed in realization
Record, since visit capacity is huge, this is all prodigious challenge to operational efficiency and storage overhead.It is efficient that the application uses support
Redis is read and write to store exposure history, application server is the one media content queue accessed of each user maintenance, when
When user uses applications client browse for media content, the behavior of user's browse for media content is reported to application by applications client
Server, at this time application server judge whether the media content queue accessed corresponding with the user, if deposited
Then this access behavioral data is being saved in from the tail portion of queue in queue, is otherwise building an empty team for the user
Row, and this access behavioral data is saved in queue.It is described to access in order to solve the storage problem of stale data
The length that media content queue is kept fixed is deleted beyond the data in the length thereof from the head of queue.It is pushed to user
When associated recommendation media content, for each correlation docking media content in the associated recommendation media content, first at this
Check whether that there are the associated recommendation media contents in the media content queue of user accessed, if it is present described
Associated recommendation media content is not as the object pushed to user.It can ensure that user is reading associated recommendation media content in this way
When do not occur any media content accessed, ensure the reading experience of user.Fig. 9 is shown when media content is article
When, the influence using duplicate removal strategy to whole article clicking rate CTR and from media article (OM articles) clicking rate.Wherein on curve
The corresponding gray scale of each coordinate points is using the dispensing ratio of the associated recommendation article of duplicate removal strategy, wherein the curve of lower section is whole
The curve of article CTR, top are OM articles CTR, it can be seen that using after duplicate removal strategy, the clicking rate of article improves.Simultaneously
It finds to also increase using the article sum of duplicate removal strategy post-exposure in practice.That is, making more articles by user
It browses to, improves the information push effect of whole system.
In some instances, the media content recommendations method of the application proposition further includes:By the associated recommendation media of acquisition
The interest of content combination user is further filtered.Specifically, it is obtained according to the record of user's history access media content
Interest characteristics of the user on user's portrait match the associated recommendation media content of acquisition with the interest characteristics of user,
Obtain the matching degree between each alternative media content and user interest profile, the higher associated recommendation media content of matching degree
It indicates that user is bigger to the interested possibility of associated recommendation media content, is recommending associated recommendation media content to user
When, filter out the smaller associated recommendation media content of matching degree.
Corresponding above-mentioned media content recommendations method, some examples of the application also provide a kind of media content recommendations device,
The device can be applied to the application server 103 in pushed information platform 102, and as shown in Figure 10, described device includes:
First similarity determining unit 1001 accesses note for obtaining the user in predetermined amount of time for each media content
Record, and the first similarity between the determining each two media content of record is accessed according to the user and is stored, first phase
The similarity of the access user of two media contents is characterized like degree;
Media content recommendations request reception unit 1002, the media content recommendations for receiving applications client transmission are asked
It asks, wherein described to answer when the first media content in each media content is accessed by the user in the applications client
The media content recommendations request is sent with client;
Alternative media content determining unit 1003, for being asked in response to the media content recommendations, from the institute stored
The first similarity that each second media content and first media content are obtained in the first similarity is stated, it will be with first matchmaker
The first similarity held in vivo is more than each second media content of the first predetermined threshold value as alternative media content;
Second similarity calculated 1004, for calculating between each alternative media content and first media content
The second similarity, the content similarity of two media contents of the second similarity characterization;
Associated recommendation media content determination unit 1005, for based in each alternative media in the alternative media content
Hold recommending for the first similarity and each alternative media content of the second similarity calculation between first media content
Property score, and be more than the alternative media content of the second predetermined threshold value as the phase of first media content using recommendability score
It closes and recommends media content;
Transmission unit 1006, it is described for the link of the associated recommendation media content of first media content to be sent to
Applications client.
Using media content recommendations device provided by the present application, on the basis for obtaining alternative media content by collaborative filtering
On, user's similitude, the content further considered between each alternative media content and the media content that user browses is similar
Property to alternative media content carry out further filter after obtain associated recommendation media content, excavate based on media content more
More correlations and readable stronger media content.
In some embodiments of the present application, above-mentioned first similarity determining unit 1001 includes:
First computing module, for accessing record, meter according to the user of each media content in nearest first preset time period
The first similarity between each two media content is calculated, a period 1 is often undergone and recalculates once, and by described first
Similarity stores in the first aggregate, and the period 1 is less than first preset time;
Second computing module, for accessing record, meter according to the user of each media content in nearest second preset time period
The first similarity between each two media content is calculated, a second round is often undergone and recalculates once, and by described first
Similarity is stored in second set, and the second round is less than second preset time, and second preset time is less than
First preset time, the second round are less than the period 1;
Above-mentioned alternative media content determining unit 1003 includes:
First similarity searches acquisition module, for judging whether be stored with each second media content in the first set
With the first similarity of first media content, each second media content is obtained if so, being searched in the first set
With the first similarity of first media content;Otherwise it is searched in the second set and obtains each second media content and institute
State the first similarity of the first media content.
In some embodiments of the present application, above-mentioned associated recommendation media content determination unit 1005 is used for each candidate
First similarity of media content and the second Similarity-Weighted sum to obtain the recommendability score of each alternative media content.
In some embodiments of the present application, above-mentioned associated recommendation media content determination unit 1005 is used for:
The temperature of each alternative media content and stylish degree are obtained,
First similarity of each alternative media content, the second similarity, temperature and/or stylish degree weighted sum are obtained
The recommendability score of each alternative media content.
In some embodiments of the present application, the media content recommendations device further includes:
Duplicate removal unit, for being rejected in the associated recommendation media content in the associated recommendation media being accessed by the user
Hold.
Above-mentioned each module may be realized in the same server apparatus or server cluster, it is also possible to be distributed in difference
Server apparatus or server cluster in.
The realization principle of above-mentioned each functions of modules has been described in detail above, and which is not described herein again.
In one example, each module in above-mentioned media content recommendations device may operate in various computing devices, and add
It is loaded in the memory of the computing device.
Figure 11 shows the composite structural diagram of the computing device where media content recommendations device.As shown in figure 11, the meter
It includes one or more processor (CPU) 1102, communication module 1104, memory 1106, user interface 1110 to calculate equipment, with
And the communication bus 1108 for interconnecting these components.
Processor 1102 can send and receive data to realize network communication and/or locally lead to by communication module 1104
Letter.
User interface 1110 includes one or more output equipments 1112 comprising one or more speakers and/or one
A or multiple visual displays.User interface 1110 also includes one or more input equipments 1114 comprising such as, key
Disk, mouse, voice command input unit or loudspeaker, touch screen displays, touch sensitive tablet, posture capture camera or other are defeated
Enter button or control etc..
Memory 1106 can be high-speed random access memory, such as DRAM, SRAM, DDR RAM or other deposit at random
Take solid storage device;Or nonvolatile memory, such as one or more disk storage equipments, optical disc memory apparatus, sudden strain of a muscle
Deposit equipment or other non-volatile solid-state memory devices.
Memory 1106 stores the executable instruction set of processor 1102, including:
Operating system 1116 includes the journey for handling various basic system services and for executing hardware dependent tasks
Sequence;
Include the various application programs for media content recommendations using 1118, this application program can be realized above-mentioned
Process flow in each example, for example may include unit some or all of in media content recommendations device shown in Fig. 10.
At least one of each unit 1001-1006 units can be stored with machine-executable instruction.Processor 1102 is deposited by executing
Machine-executable instruction in reservoir 1106 in each unit 1001-1006 at least one unit, and then can realize above-mentioned each
The function of at least one of unit 1001-1006 modules.
It should be noted that step and module not all in above-mentioned each flow and each structure chart is all necessary, it can
To ignore certain steps or module according to the actual needs.Each step execution sequence be not it is fixed, can as needed into
Row adjustment.The division of each module is intended merely to facilitate the division functionally that description uses, and in actual implementation, a module can
It is realized by multiple modules with point, the function of multiple modules can also be realized by the same module, these modules can be located at same
In a equipment, it can also be located in different equipment.
Hardware module in each embodiment can in hardware or hardware platform adds the mode of software to realize.Above-mentioned software
Including machine readable instructions, it is stored in non-volatile memory medium.Therefore, each embodiment can also be presented as software product.
In each example, hardware can be by special hardware or the hardware realization of execution machine readable instructions.For example, hardware can be with
Permanent circuit or logical device (such as application specific processor, such as FPGA or ASIC) specially to design are used to complete specifically to grasp
Make.Hardware can also include programmable logic device or circuit by software provisional configuration (as included general processor or other
Programmable processor) for executing specific operation.
In addition, each example of the application can pass through the data processor by data processing equipment such as computer execution
To realize.Obviously, data processor constitutes the application.In addition, being generally stored inside the data processing in a storage medium
Program by program by directly reading out storage medium or the storage by program being installed or being copied to data processing equipment
It is executed in equipment (such as hard disk and/or memory).Therefore, such storage medium also constitutes the application, and present invention also provides one
Kind non-volatile memory medium, wherein being stored with data processor, this data processor can be used for executing in the application
State any one of method example example.
The corresponding machine readable instructions of module in Figure 11 can make operating system operated on computer etc. complete this
In some or all of operation that describes.Non-volatile computer readable storage medium storing program for executing can be the expansion board being inserted into computer
In in set memory or write the memory being arranged in the expanding element being connected with computer.Mounted on expansion board
Or CPU on expanding element etc. can be according to instruction execution part and whole practical operations.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
With within principle, any modification, equivalent substitution, improvement and etc. done should be included within the scope of protection of the invention god.
Claims (12)
1. a kind of media content recommendations method, which is characterized in that including:
User's access record for each media content in predetermined amount of time is obtained, and record is accessed according to the user and is determined often
The first similarity between two media contents simultaneously stores, the access user's of first similarity characterization, two media contents
Similarity;
Receive the media content recommendations request that applications client is sent, wherein when in the first media in each media content
When appearance is accessed by the user in the applications client, the applications client sends the media content recommendations request;
In response to the media content recommendations ask, obtained from first similarity stored each second media content with
First similarity of first media content will be more than the first predetermined threshold value with the first similarity of first media content
Each second media content as alternative media content;
Calculate the second similarity between each alternative media content and first media content, second similarity characterization
The content similarity of two media contents;
Based on the first similarity in the alternative media content between each alternative media content and first media content
And second each alternative media content of similarity calculation recommendability score, and be more than the second default threshold by recommendability score
Associated recommendation media content of the alternative media content of value as first media content;
The link of the associated recommendation media content of first media content is sent to the applications client.
2. media content recommendations method according to claim 1, wherein for each media in the acquisition predetermined amount of time
The user of content accesses record, and accesses the first similarity packet between the determining each two media content of record according to the user
It includes:
It is accessed and is recorded according to the user of each media content in nearest first preset time period, between calculating each two media content
First similarity often undergoes a period 1 and recalculates once, and in the first aggregate by first similarity storage,
The period 1 is less than first preset time;
It is accessed and is recorded according to the user of each media content in nearest second preset time period, between calculating each two media content
First similarity often undergoes a second round and recalculates once, and first similarity is stored in second set,
The second round is less than second preset time, and second preset time is less than first preset time, and described the
Two cycles are less than the period 1;
It is described to obtain each second media content and the first similarity of first media content includes:
Judge the first similarity that each second media content and first media content whether are stored in the first set,
If so, searching the first similarity for obtaining each second media content and first media content in the first set;
Otherwise the first similarity for obtaining each second media content and first media content is searched in the second set.
3. media content recommendations method according to claim 2, wherein the calculated between each two media content
One similarity includes:
Obtain the access user vector of each media content in described two media contents;
Calculate the cosine similarity of the access user vector of described two media contents;
Using the cosine similarity being calculated as the first similarity between described two media contents.
4. media content recommendations method according to claim 1, wherein it is described calculate each alternative media content with it is described
The second similarity between first media content includes:
Obtain the keyword vector of each alternative media content;
Obtain the keyword vector of first media content;
The keyword vector for calculating first media content is similar to each cosine of keyword vector of alternative media content
Degree;
Using the cosine similarity being calculated as between each alternative media content and first media content
Two similarities.
5. media content recommendations method according to claim 1, wherein described to calculate pushing away for each alternative media content
The property recommended score includes:First similarity of each alternative media content and the second Similarity-Weighted are summed to obtain each candidate matchmaker
The recommendability score held in vivo.
6. media content recommendations method according to claim 1, wherein described to calculate pushing away for each alternative media content
The property recommended score includes:
The temperature of each alternative media content and stylish degree are obtained,
First similarity of each alternative media content, the second similarity, temperature and/or stylish degree weighted sum are obtained each
The recommendability score of alternative media content.
7. media content recommendations method according to claim 1, wherein the method further includes:
The associated recommendation media content being accessed by the user is rejected in the associated recommendation media content.
8. a kind of media content recommendations device, which is characterized in that including:
First similarity determining unit accesses record, and root for obtaining the user in predetermined amount of time for each media content
The first similarity between the determining each two media content of record is accessed according to the user and is stored, first similarity characterization
The similarity of the access user of two media contents;
Media content recommendations request reception unit, the media content recommendations request for receiving applications client transmission, wherein when
When the first media content in each media content is accessed by the user in the applications client, the applications client hair
The media content recommendations are sent to ask;
Alternative media content determining unit, for being asked in response to the media content recommendations, from first phase stored
Like the first similarity for obtaining each second media content and first media content in degree, by with first media content
First similarity is more than each second media content of the first predetermined threshold value as alternative media content;
Second similarity calculated, for calculating the second phase between each alternative media content and first media content
Like degree, the content similarity of two media contents of the second similarity characterization;
Associated recommendation media content determination unit, for based on each alternative media content in the alternative media content with it is described
The recommendability score of each alternative media content of the first similarity and the second similarity calculation between first media content, and
It is more than the alternative media content of the second predetermined threshold value as the associated recommendation matchmaker of first media content using recommendability score
Hold in vivo;
Transmission unit, for the link of the associated recommendation media content of first media content to be sent to the application client
End.
9. media content recommendations device according to claim 8, wherein first similarity determining unit includes:
First computing module calculates every for accessing record according to the user of each media content in nearest first preset time period
The first similarity between two media contents often undergoes a period 1 and recalculates once, and similar by described first
In the first aggregate, the period 1 is less than first preset time for degree storage;
Second computing module calculates every for accessing record according to the user of each media content in nearest second preset time period
The first similarity between two media contents often undergoes a second round and recalculates once, and similar by described first
Degree is stored in second set, and the second round is less than second preset time, and second preset time is less than described
First preset time, the second round are less than the period 1;
The alternative media content determining unit includes:
First similarity searches acquisition module, for judging whether be stored with each second media content and institute in the first set
The first similarity of the first media content is stated, each second media content and institute are obtained if so, being searched in the first set
State the first similarity of the first media content;Otherwise it is searched in the second set and obtains each second media content and described the
First similarity of one media content.
10. media content recommendations device according to claim 8, wherein the associated recommendation media content determination unit,
For the first similarity of each alternative media content and the second Similarity-Weighted to be summed to obtain each alternative media content
Recommendability score.
11. media content recommendations device according to claim 8, wherein the associated recommendation media content determination unit,
For:
The temperature of each alternative media content and stylish degree are obtained,
First similarity of each alternative media content, the second similarity, temperature and/or stylish degree weighted sum are obtained each
The recommendability score of alternative media content.
12. media content recommendations device according to claim 8, wherein described device further includes:
Duplicate removal unit, for rejecting the associated recommendation media content being accessed by the user in the associated recommendation media content.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710037620.6A CN108319622B (en) | 2017-01-18 | 2017-01-18 | Media content recommendation method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710037620.6A CN108319622B (en) | 2017-01-18 | 2017-01-18 | Media content recommendation method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108319622A true CN108319622A (en) | 2018-07-24 |
CN108319622B CN108319622B (en) | 2022-11-11 |
Family
ID=62892040
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710037620.6A Active CN108319622B (en) | 2017-01-18 | 2017-01-18 | Media content recommendation method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108319622B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111046230A (en) * | 2019-12-06 | 2020-04-21 | 北京奇艺世纪科技有限公司 | Content recommendation method and device, electronic equipment and storable medium |
CN112256970A (en) * | 2020-10-28 | 2021-01-22 | 四川金熊猫新媒体有限公司 | News text pushing method, device, equipment and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103106208A (en) * | 2011-11-11 | 2013-05-15 | 中国移动通信集团公司 | Streaming media content recommendation method and system in mobile internet |
CN105868248A (en) * | 2015-12-15 | 2016-08-17 | 乐视网信息技术(北京)股份有限公司 | Media recommendation method and device |
CN106126669A (en) * | 2016-06-28 | 2016-11-16 | 北京邮电大学 | User collaborative based on label filters content recommendation method and device |
-
2017
- 2017-01-18 CN CN201710037620.6A patent/CN108319622B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103106208A (en) * | 2011-11-11 | 2013-05-15 | 中国移动通信集团公司 | Streaming media content recommendation method and system in mobile internet |
CN105868248A (en) * | 2015-12-15 | 2016-08-17 | 乐视网信息技术(北京)股份有限公司 | Media recommendation method and device |
CN106126669A (en) * | 2016-06-28 | 2016-11-16 | 北京邮电大学 | User collaborative based on label filters content recommendation method and device |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111046230A (en) * | 2019-12-06 | 2020-04-21 | 北京奇艺世纪科技有限公司 | Content recommendation method and device, electronic equipment and storable medium |
CN111046230B (en) * | 2019-12-06 | 2023-09-15 | 北京奇艺世纪科技有限公司 | Content recommendation method and device, electronic equipment and storable medium |
CN112256970A (en) * | 2020-10-28 | 2021-01-22 | 四川金熊猫新媒体有限公司 | News text pushing method, device, equipment and storage medium |
CN112256970B (en) * | 2020-10-28 | 2024-06-25 | 四川金熊猫新媒体有限公司 | News text pushing method, device, equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN108319622B (en) | 2022-11-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11269962B2 (en) | Inductive matrix completion and graph proximity for content item recommendation | |
JP5881153B2 (en) | Recommendation of correlated information | |
US20190197416A1 (en) | Information recommendation method, apparatus, and server based on user data in an online forum | |
US7519588B2 (en) | Keyword characterization and application | |
JP6167493B2 (en) | Method, computer program, storage medium and system for managing information | |
JP2015191655A (en) | Method and apparatus for generating recommendation page | |
JP2014522004A (en) | Send product information based on determined preference values | |
EP2941724A1 (en) | Method and apparatus for generating webpage content | |
CN109325179A (en) | A kind of method and device that content is promoted | |
TW200907718A (en) | Dynamic layout for a search engine results page based on implicit user feedback | |
JP2014501013A (en) | Product information ranking | |
US10019419B2 (en) | Method, server, browser, and system for recommending text information | |
CN105095279B (en) | File recommendation method and device | |
CN108363815A (en) | A kind of pre-reading method of Webpage, device and intelligent terminal | |
WO2018205845A1 (en) | Data processing method, server, and computer storage medium | |
CN108416006A (en) | Information off-line recommends method, apparatus, client, server and storage medium | |
CN106156351B (en) | Multimedia resource recommendation information generation method and device | |
JP2012501492A (en) | Method and system for providing product object information on the Internet | |
CN111476469A (en) | Guest expansion method, terminal equipment and storage medium | |
CN108021619A (en) | A kind of event description object recommendation method and device | |
CN109522473B (en) | Method for recommending associated information, terminal and server thereof | |
CN109688178A (en) | Recommended method, device and equipment | |
KR101307105B1 (en) | Information provisioning device, information provisioning method, and information recording medium | |
CN108319622A (en) | A kind of media content recommendations method and device | |
CN111444447A (en) | Content recommendation page display method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |