CN103974097B - Personalized user original video forecasting method based on popularity and social networkies and system - Google Patents

Abstract

The invention discloses a personalized user original video prefetching method and system based on popularity and social network. This method first collects video popularity information, social relations between users, association relations between videos, and user historical behavior information; based on the popularity information, user-created videos are divided into popular videos and long-tail videos; Prefetch list of popular videos is generated by sorting preferences; build a graph model to measure the correlation between users and videos, and generate long-tail video pre-fetch lists for users; weighted linear fusion of popular video pre-fetch lists and long-tail video pre-fetch lists , build a personalized hybrid prefetch model; in the case of ensuring that the user is currently watching the video smoothly, based on the personalized hybrid prefetch model, prefetch one or more top-ranked videos for the user. This system improves the hit rate and accuracy of user original video prefetching, improves the quality of user original video service, and improves the user's viewing experience.

Description

Personalized user-generated video prefetching method based on popularity and social network and its system

technical field

The invention relates to a personalized user original video prefetching method and system based on popularity and social network, and belongs to the technical field of video prefetching.

Background technique

With the advent of the Web 2.0 era, users are no longer simply receivers of information, but have become publishers of information at the same time. Represented by Youku and Tudou, video sites based on user-generated content are popular among users. Highly sought after, but multiple research results show that the user's viewing experience is unsatisfactory, and the delay caused by playback accounts for a relatively high proportion of the video duration. Related methods to reduce video playback delay include: improving server hardware and software performance, increasing network bandwidth, based on network proxy and based on content distribution network, but these methods have corresponding disadvantages, compared with video prefetching technology can be very good It reduces the user's perception of delay, and at the same time generates a small overhead.

However, since video services based on user-generated content have many characteristics different from traditional video services in terms of user behavior and video content, such as a large number of videos, short video length, less content data, and extremely fast content generation speed, different Popularity distribution and existence of social network. Therefore, the traditional video prefetching algorithm is not well suited for user-generated content video services.

The popularity distribution of user-generated videos is extremely unbalanced. Some videos have high popularity. When choosing to watch these videos, users usually consider the popularity of the videos and whether they meet their own interests. Another part of the videos is not very popular, but they are watched by interested users through different transmission channels. The main transmission channels include the social relationship between users and the association relationship between videos.

Therefore, it is necessary to propose a video prefetching method and system suitable for user-created videos that can comprehensively consider videos of different popularity, make full use of user social relations, video association relations, video popularity information, and user historical behavior data. Effectively improve the hit rate and accuracy of video prefetching, improve the quality of user-original video services, and improve the user's viewing experience.

Contents of the invention

The technical problem to be solved by the present invention is to provide a personalized user original video prefetching method and system based on popularity and social network. The method and system improve the hit rate and accuracy of video prefetching, improve the quality of user original video services, and improve the user's viewing experience.

In order to achieve the above object, according to an aspect of the present invention, the present invention provides a method for prefetching user-generated videos based on popularity and social networks, comprising the following steps:

1) Information Acquisition: Obtain the user's historical behavior information on the video, the social relationship between the user and the user, the association between the video and the popularity information of the original video;

2) Video classification: Based on the popularity information of user-generated videos, videos are divided into two categories: popular videos and long-tail videos;

3) Popular video prefetch list generation: based on the popularity information of original videos and user historical behavior information, calculate the user's preference for popular videos, sort them in descending order according to preference, and generate a popular video prefetch list for users;

4) Graph model construction: Construct a graph model based on the obtained user historical behavior information, social relationships between users, and associations between videos, in which each node or edge has a node or relationship type that is consistent with the actual physical type, Each edge has an edge weight consistent with the actual relationship strength;

5) Long tail video prefetch list generation: Based on the graph model, using the correlation degree measurement method based on the shortest path weight, measure the correlation between users and videos, sort according to the descending order of correlation, and generate long tail video prefetch lists for users;

6) Hybrid prefetching model generation: Based on the acquired user's historical behavior information, the popular video prefetching list and the long tail video prefetching list are weighted and linearly fused to build a personalized hybrid prefetching model;

7) Generate a prefetch video list: In the case of ensuring that the user is currently watching the video smoothly, based on the personalized hybrid prefetch model, prefetch one or more video recommendations that are ranked first for the user.

In the above user original video prefetching method, the information acquisition step includes: if there is a social relationship between users or an association relationship between videos in the application system, then acquire these information, if not, then No need to obtain; the user's historical behavior information on the video includes the user's behavior of watching, sharing, uploading or favoriteing the video.

In the above user original video prefetching method, before the step of generating the popular video prefetch list, it also includes: a preprocessing step for user historical behavior data, specifically: only keep the information that the user affirms the video; , including the user's viewing, sharing, collection and uploading information of the video.

In the above user-generated video prefetching method, the video classification steps are as follows: firstly, use the number of video views to measure the popularity of the video, and sort the videos according to the popularity from high to low; secondly, the cumulative number of views accounts for 80% of the total number of views %, the top 10% of the videos are popular videos, if the cumulative views of the top 10% of the videos are less than 80% of the total views, the popular videos will be expanded backwards according to the popularity ranking until the cumulative views account for 80% of the total views. 80% of the total views; finally, all popular videos are excluded, and the remaining videos are long-tail videos.

In the above-mentioned user original video prefetching method, in the popular video prefetching list generation step, the specific process of preference calculation is as follows: first, use the video viewing times to calculate video popularity; second, use user historical behavior information to calculate user The interest factor of the video; finally, the preference of the user's video is calculated by combining the popularity and the interest factor.

In the above-mentioned user-generated video prefetching method, the graph model construction steps are as follows: first, map the user entity and the video entity to corresponding nodes in the graph, and use the type of the entity as the type of the corresponding node; secondly, if If there is a social relationship between users, add an edge between the corresponding nodes, and the weight of the edge represents the strength of the social relationship. If there is an association between the videos, add an edge between the corresponding nodes, and the weight of the edge represents the association Relationship strength; finally, if the user has affirmative behavior information on the video, an edge is added between the user and the video, and the weight of the edge represents the strength of the affirmative behavior.

In the above user original video prefetching method, in the long-tail video prefetching list generation step, the specific process of correlation calculation is as follows: first, calculate the shortest path between nodes, if there is no path between nodes, The correlation degree is 0; secondly, the weight of a single shortest path is calculated; finally, the sum of the weights of the shortest path is used as the correlation between nodes.

In the above user-generated video prefetching method, the hybrid prefetching model generation step specifically includes: using the user's historical behavior information, respectively calculating the proportion of the user's behaviors on popular videos and long-tail videos to the total number of behaviors, and calculating The result is used as the weight of the popular video prefetch list and the long tail video prefetch list, and the weighted linear fusion is performed on the popular video prefetch list and the long tail video prefetch list.

In addition, the present invention provides a personalized user-original video prefetching system based on popularity and social attributes, including: an information acquisition module for acquiring historical behavior information of users on videos, social relations between users, The association relationship between videos and the popularity information of the video itself; the preprocessing module is used to retain positive information; the video classification module is used to divide videos into popular videos and long-form videos based on the popularity information of user-created videos There are two types of tail videos; the popular video prefetch list generation module is used to calculate the user's preference for popular videos based on video popularity and user historical behavior information, and generate a popular video prefetch list; the graph model building module is used to Construct a graph model containing node types and edge weights for the historical behavior information of the video, the social relationship between users and the relationship between the videos; the long-tail video prefetch list generation module is used for the graph-based model, using the shortest path-based The weight correlation measurement method calculates the correlation between users and long-tail videos, and generates long-tail video prefetch lists; the hybrid prefetch model generation module is used to calculate popular video prefetch weights and long-tail video based on user historical behavior information. Tail video prefetch weight, weighted linear fusion popular video prefetch list and long tail video prefetch list, generate a personalized hybrid prefetch model; Personalize the hybrid model to prefetch the top-ranked video or videos for the user.

The present invention proposes a personalized user original video prefetching method and system based on popularity and social network. For the first time, based on the popularity distribution of user-generated videos, videos are divided into two categories: popular videos and long-tail videos. Based on the user's historical behavior data, the user's preference for unwatched videos is mined, video prefetch lists are generated for the two types of videos, and the two types of video prefetch lists are fused linearly through the hybrid prefetch model. Therefore, the hit rate and accuracy of video prefetching are effectively improved, the quality of user original video services is improved, and the viewing experience of users is improved.

Description of drawings

Fig. 1 is the flow chart of the steps of the personalized user original video prefetching method based on popularity and social network in the present invention;

FIG. 2 is a schematic diagram of a scene description according to an embodiment of the present invention;

Fig. 3 is a schematic diagram of a graph model of an embodiment of the present invention;

FIG. 4 is a schematic diagram of video prefetching according to an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of the personalized user-original video prefetching system based on popularity and social network of the present invention.

detailed description

The invention provides a personalized user original video prefetching method and system based on popularity and social network. A detailed description will be given below in conjunction with the accompanying drawings.

With reference to Fig. 1, Fig. 1 is the flow chart of the steps of the personalized user original video prefetching method based on popularity and social attributes of the present invention, including the following steps:

Information obtaining step 101: obtaining the user's historical behavior information on the video, the social relationship between users, the association relationship between videos and the popularity information of the video itself;

Video classification step 102: based on the popularity information of user-generated videos, classify user-generated videos into popular videos and long-tail videos;

Popular video prefetch list generation step 103: based on video popularity and user's historical behavior information to video, calculate user's preference degree to popular video, sort according to preference degree descending order, generate popular video prefetch list for user;

Graph model construction step 104: construct a graph model based on the user’s historical behavior information on videos, the social relationship between users and the association relationship between videos, and each node or edge has a node or relationship type that is consistent with the actual physical type , each edge has an edge weight consistent with the actual relationship strength;

Long-tail video prefetch list generation step 105: Based on the graph model, use the correlation degree measurement method based on the shortest path weight to measure the correlation between users and videos, sort according to the descending order of correlation, and generate long-tail video prefetch lists for users;

Step 106 of generating a hybrid prefetching model: based on the acquired historical behavior information of the user, weighted linear fusion is performed on the popular video prefetching list and the long tail video prefetching list to construct a personalized hybrid prefetching model;

Video prefetching step 107: under the condition that the user is currently watching the video smoothly, based on the personalized hybrid prefetching model, prefetch one or more video recommendations that are ranked the highest for the user.

A preferred processing method in practice is to include a preprocessing step before the popular video prefetch list generation step, process the user's historical behavior information, and only keep the information that the user affirms the video; in the non-scoring system, including user Watch, share, bookmark, and upload information about videos.

The specific implementation of this method will be described in detail below in conjunction with an example of a user-generated video service platform. With reference to Fig. 2, Fig. 2 is the scene description diagram of this embodiment, and this embodiment has described a user original video service platform, and video has its own content information; Social relations can be established between users; Users can watch, bookmark, upload, Share videos; there is an association between videos, which is added by users when uploading videos.

According to the user's behavior on the video, the user-video relationship matrix R can be constructed. If user i has watched, collected, uploaded or shared the video j, R _ij is 1, otherwise R _ij is 0. R is an m×n-dimensional matrix, m is the number of users in the system, and n is the number of videos in the system.

Classify videos according to the popularity distribution: First, use the number of video views to measure the popularity of videos, as shown in formula (1), where P _k represents the popularity of video k, and M _k represents the number of times video k is viewed. Secondly, according to the video popularity from high to low, the accumulated views account for 80% of the total views, and the top 10% of the videos are popular videos. If the cumulative views of the top 10% of the videos are less than the total views 80% of the popular videos are expanded backwards according to the popularity ranking until the cumulative views account for 80% of the total views; finally, all popular videos are excluded, and the remaining videos are long-tail videos. represents a collection of popular videos, Represents a collection of long-tail videos.

P _k =log(M _k +1) (1)

The following describes the method for calculating the user's preference for popular videos during the process of generating the popular video prefetch list. First, use formula (1) to calculate the popularity P _k of video k; secondly, calculate the interest factor I _i (k) of user i to video k, specifically as formula (2), where Indicates that user i has a collection of affirmative action videos, Represents the collection of N videos most similar to video k, and Sim(k,j) represents the similarity between video k and video j.

The similarity is calculated using the cosine similarity, as shown in formula (3), where N(k) represents the set of users who like video k, and is normalized by the maximum value, as shown in formula (4).

Finally, formula (5) is used to calculate user i's preference S _i (k) for popular video k.

S _i (k) = P _k × I _i (k) (5)

Next, build a graph model based on the user-video relationship matrix R, the social relationship between users and the relationship between videos. A graph model refers to an undirected graph G(U, V, E, R, W; Φ) with sideband weights. Among them, U represents the user collection, including all user nodes; V represents the video collection, including all video nodes; E represents the set of edges in the graph, including all relationships between nodes in the social network and all user affirmative behaviors; R represents A collection of relationship types, including all relationship types; Φ:E→R, representing the mapping function of relationship types; W represents the weight of edges in graph G, which measures the strength of a certain relationship; all edges e in E belong to A type of relationship, Φ(e)∈R; e(u,v)∈E, u,v∈U|V, represents a certain type of relationship between node u and node v; and w(e) represents The weight of the edge e; where |R|>1, when |R|=1, G degenerates into a user-item bipartite graph.

According to the definition of the graph model, the specific construction process is as follows:

1) user entities and video entities are mapped to corresponding nodes in graph G, and the type of the entity is used as the type of the corresponding node;

2) If there is a social relationship between user u _i and user u _j , add an edge e(u _i , u _j ), and the weight ω(e) of the edge represents the strength of the social relationship, which is calculated using formula (6), where F _i represents the friend set of user i, and F _j represents the friend set of user j;

3) If there is an association relationship between video v _i and video v _j , then add edge e(v _i , v _j ), and the weight ω(e) of the edge represents the strength of the association relationship, which is calculated by formula (7), where represents the collection of related videos for video i, Represents the collection of related videos for video j. If video v _j is also a related video of video v _i , since it is an undirected graph, edge e(v _j ,v _i ) already exists, so there is no need to add a new edge, just update the weight of the edge to the original double;

4) If user u _i has affirmative behavior on video v _j , then add edge e(u _i , v _j ), and edge weight ω(e) is 1/n, representing the strength of affirmative behavior, where n is user u _iTotal number of videos with affirmative action

Referring to Fig. 3, the method for calculating the correlation between user u _i and long tail video v _j is introduced based on the graphical model in the process of generating the long tail video prefetch list:

1) In G, judge whether there is a path between node u _i and node v _j , if not, the similarity is 0; if yes, calculate all the shortest paths between node u _i and node v _j , define Γ (u _i , v _j ) is the collection of all shortest paths between node u _i and node v _j in graph G. Although G is a graph with edge weights, when calculating the shortest path between nodes, the weight of edges between nodes is not considered, only the number of edges on the path is considered;

2) For the shortest path ρ∈Γ(u _i ,v _j ) between node u _i and node v _j , calculate its weight ω(ρ). Assuming the shortest path ρ={u _i ,...,v _j }, ω(ρ) can be calculated according to formula (8), using {n ₁ ,...,n _k } to represent the nodes passed by the shortest path ρ , n ₁ is u _i , n _k is v _j . ω(n _i ,n _i+1 ) represents the weight of edge e(n _i ,n _i+1 ), and out(n _i ) represents the degree of node n _i . ;

3) For all the shortest paths ρ∈Γ(u _i , v _j ) between node u _i and node v _j , add their weights ω(ρ) to calculate the correlation between node u _i and node v _j sim(u _i , v _j ), as shown in formula (9);

The steps for building a personalized hybrid prefetching model are described below. Since it is a personalized hybrid prefetching model, it is necessary to fully consider the characteristics of different users. Taking user u _i as an example, use user u _i 's historical behavior information on videos to calculate the popular video prefetch list The weight ω _P (u _i ) and the long tail video prefetch list The weight ω _L (u _i ), through the linear weighted fusion method, fuses the popular video prefetch list and the long tail video prefetch list to generate a personalized hybrid prefetch model The details are as formula (10).

The calculation of popular video prefetch list weight ω _P (u _i ) is specifically as formula (11), and the calculation of long tail video prefetch list weight ω _L (u _i ) is specific as formula (12).

Next, refer to Figure 4 to introduce the steps of video prefetching:

1) The video cache is divided into two parts, the playback cache and the prefetch cache. The playback cache is used to cache the data blocks of the video content currently being watched, and the prefetch cache is used to cache the header data blocks of the prefetched video content;

2) Receive a personalized hybrid prefetch model from the server, and use this hybrid prefetch model to predict the video content that the user may watch in the future during the playback process;

3) The data scheduling strategy includes two stages. In each scheduling cycle, the urgent data block is first downloaded to the playback cache using the download bandwidth to ensure that the video currently being watched by the user can be played smoothly. The urgent data block refers to the Playing order, the data blocks whose playing time is close to the current playing time; if there is still ample download bandwidth after downloading the urgent data block, predict the video that the user may watch in the future based on the hybrid prefetching model, and use the remaining download bandwidth to upload the title of the video Content, prefetched into the prefetch cache.

Finally, a personalized user-generated video prefetching system based on popularity and social network is described with reference to FIG. 5 . As shown in Figure 5, the system specifically includes: an information acquisition module 501, which is used to acquire the user's historical behavior information on the video, the social relationship between users, the association relationship between videos and the popularity of the video itself Information; preprocessing module 502, for retaining positive information; video classification module 503, for distributing the popularity based on user original video, video is divided into popular video and long-tail video two classes; popular video prefetching list generation module 504, for calculating the user's preference for popular videos based on video popularity and user historical behavior information, and generating a popular video prefetch list; graph model building module 505, for based on the user's historical behavior information for videos, between users Build a graph model containing node types and edge weights between social relationships and video associations; the long-tail video prefetch list generation module 506 is used to based on the social relationship between users, the association relationship between videos, and user history behavior information , calculate the correlation between the user and the long-tail video, and generate the long-tail video prefetch list; the hybrid prefetch model generation module 507 is used to calculate the popular video prefetch weight and the long-tail video prefetch weight based on the user's historical behavior information , weighted linear fusion popular video prefetch list and long-tail video prefetch list, generate personalized hybrid prefetch model; video prefetch module 508, for ensuring that the user is currently watching the video smoothly, based on the personalized hybrid model, Prefetch the top-ranked video or videos for the user

Thus, the personalized user original video prefetching system based on the popularity and social network of the embodiment of the present invention is realized.

It is explained here that the principle of the system embodiment is the same as that of the method embodiment, and they can be referred to each other. I won't repeat them here.

Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, they should not be construed as limiting the present invention. It should be pointed out that those skilled in the art can make some modifications and variations without departing from the principle and scope of the present invention, and these modifications and variations should also be considered within the protection scope of the present invention.

Claims (11)

1. a kind of personalized user original video forecasting method based on popularity and social networkies it is characterised in that include with Lower step：

1) acquisition of information:Obtain the historical behavior information to video for the user, the social networks between user and user, video with regard Incidence relation between frequency and the popularity information of original video；

2) visual classification：Based on the popularity information of the original video obtaining, video is divided into popular video and long-tail video two Class；

3) generate popular video and prefetch list：Popularity information based on the original video obtaining and user's history behavioural information, Calculate the preference to popular video for the user, according to preference descending sort, be that the popular video of user's generation prefetches list；

4) graph model builds：Based on the user's history behavioural information obtaining, between the social networks between user and video Incidence relation builds graph model, and each node or every a line have the node type consistent with actual physics type Or relationship type, every a line has the side right value consistent with actual relationship intensity；

5) generate long-tail video and prefetch list：Based on graph model, using the degree of association measuring method based on shortest path weights, survey Amount user and the degree of association of video, according to degree of association descending sort, are that user's generation long-tail video prefetches list；

6) build personalized mixing Prefetching Model：Based on the historical behavior information of the user obtaining, list is prefetched to popular video Prefetch list with long-tail video and be weighted linear fusion, build personalized mixing Prefetching Model；

7) video prefetches：In the case of ensureing user's currently viewing video smoothness, based on personalized mixing Prefetching Model, it is use Family prefetches and comes the most forward one or multi-section video.

2. personalized user original video forecasting method according to claim 1 is it is characterised in that step 1) information obtain During taking, for the incidence relation between the social relations between user and user and video and video, if in application system Exist in system, then obtain these information；If no, need not obtain.

3. personalized user original video forecasting method according to claim 1 and 2 is it is characterised in that in described step 3), before popular video prefetches list generation, also include the step that pretreatment is carried out to user's history behavioural information, specially：Only Retain user's information to video certainly；In non-marking system, including user to the viewing of video, share, collect and on Biography information.

4. personalized user original video forecasting method according to claim 3 is it is characterised in that described step 2) video Classification detailed process be：

2.1) behavior to video according to user, build user-video relationship matrix R, if user i video j was had viewing, Collection, upload or splitting glass opaque, R_ijFor 1, otherwise R_ijIt is the matrix of m × n dimension for 0, R, m is number of users in system, N is number of videos in system；

2.2) classified according to popular Degree distributions video：First, weigh video popularity using video-see number of times, specifically As formula (1),

P_k=log (M_k+1) (1)

Wherein P_kRepresent the popularity of video k, M_kRepresent the number of times that video k is watched；

Secondly, sort from high to low according to video popularity, accumulative viewing number of times accounts for total viewing number of times 80%, before popularity rankings 10% video is popular video, if 10% video adds up to watch number of times that number of times is not watched by Football Association before popularity rankings 80%, extend popular video backward according to popularity rankings, until meeting accumulative viewing number of times to account for total viewing number of times 80%；? Afterwards, exclusion all hot topics video, remaining video is long-tail video,Represent popular video collection,Represent long-tail video collection.

5. the personalized user original video forecasting method according to claim 4 is it is characterised in that step 3) hot topic regards Frequency prefetches in list generation, and the detailed process that described preference calculates is：

3.1) the popularity P according to video k_k, calculate the interest factor I to video k for the user i_i(k), concrete such as formula (2),

WhereinRepresent that user i had the set of behavior video certainly,Represent and the most like N portion video of video k Set, Sim (k, j) represent video k and video j similarity；

3.2) similarity is calculated using cosine similarity, concrete such as formula (3), and wherein N (k) represents the use liking video k Family is gathered, and is normalized by maximum, concrete such as formula (4)

$Sim(k,j)=\frac{|N\left(k\right)\∩N\left(j\right)|}{\sqrt{\left|N\left(k\right)\right|\×\left|N\left(j\right)\right|}}---\left(3\right)$

${\mathrm{Sim}}^{\′}(k,j)=\frac{Sim(k,j)}{\underset{j}{max}Sim(k,j)}---\left(4\right)$

Formula (5) is finally utilized to calculate preference S to popular video k for the user i_i(k)

S_i(k)=P_k×I_i(k) (5).

6. personalized user original video forecasting method according to claim 5 is it is characterised in that step 4) graph model structure The detailed process built is：

First, user subject and video entities are mapped as the corresponding node of in figure, and are saved as corresponding with the type of entity The type of point；Secondly, if there are social networks between user, add side between corresponding node, the weights on side represent society Handing over relationship strength, if there is incidence relation between video, adding side between corresponding node, the weights on side represent association and close It is intensity；Finally, if user has behavioural information certainly to video, add side between user and this video, the weights on side represent Certainly the intensity of behavior.

7. personalized user original video forecasting method according to claim 6 is it is characterised in that this graph model is one Non-directed graph G (U, V, E, R, the W of sideband weights；Φ), wherein U represents user's set, including all of user node；V represents and regards Frequency is gathered, including all of video node；The set on side in E representative graph, including all of relation between social networkies interior joint And all of user behavior certainly；R representation relation type set, including all of relationship type；Φ:E → R, representation relation The mapping function of type；In W representative graph G, the weights on side, have weighed the intensity of certain relation；In E, all of side e, broadly falls into one Plant relationship type, Φ (e) ∈ R；E (u, v) ∈ E, u, v ∈ U | V, represent and between node u and node v, there is certain type of pass System；And w (e) represents the weights of this line of e；Wherein | R | ＞ 1, when | R |=1, G deteriorates to user-article two Component.

8. personalized user original video forecasting method according to claim 7 is it is characterised in that step 5) long-tail video Prefetch in list generation, the detailed process of described relatedness computation is：

First, it is calculated the shortest path between node, if there is not path between node, degree of association is 0；Secondly, calculate The weights of wall scroll shortest path；Finally, using shortest path weights sum as the degree of association between node.

9. personalized user original video forecasting method according to claim 8 is it is characterised in that step 6) mixing prefetches Model generate detailed process be：

Based on user's history behavioural information, calculating user respectively and the behavior quantity of popular video and long-tail video is accounted for head office is number The ratio of amount, result of calculation is prefetched list as popular video and long-tail video prefetches the weights of list, weighted linear merges Popular video prefetches list and long-tail video prefetches list.

10. personalized user original video forecasting method according to claim 9 is it is characterised in that step 7) video preprocessor The detailed process taking is：

1) video cache is divided into two parts, plays caching and prefetch caching, broadcasting caching is used for caching currently to be seen See the data block of video content, prefetch caching for the head data block of cache prefetching video content；

2) accept personalized mixing Prefetching Model at server, utilize this to mix Prefetching Model in playing process afterwards The following video content that may watch of prediction user；

3) data dispatch strategy includes two stages, within each dispatching cycle, first with download bandwidth download emergency data block To playing caching, to ensure that the video that user is currently viewed is capable of smooth broadcasting, emergency data block refers to according to video Playing sequence, reproduction time closes on the data block of current play time；If still having more than needed after having downloaded emergency data block Download bandwidth, based on mixing Prefetching Model predict user following may viewing video, and will using remaining download bandwidth The head content of video, is prefetched to and prefetches in caching.

A kind of 11. personalized user original video pre-fetching systems based on video popularity and social networkies are it is characterised in that institute The system of stating includes：

Data obtaining module, for obtaining the historical behavior information to video for the user, the social networks between user and user, depending on Incidence relation between frequency and video and the popularity information of video itself；

Visual classification module, for the popularity information based on user's original video, video is divided into popular video and long-tail to regard Frequently two class；

Pretreatment module, for filtering to user's history behavioural information, retains information certainly；

Popular video prefetches List Generating Module, for based on video popularity and user's history behavioural information, calculating user couple The preference of popular video, generates popular video and prefetches list；

Graph model builds module, for the historical behavior information to video based on user, social networks between user and regarding Between frequency, incidence relation builds the graph model containing node type and side right value；

Long-tail video prefetches List Generating Module, for based on graph model, using the degree of association measurement based on shortest path weights Method, calculates the degree of association of user and long-tail video, generates long-tail video and prefetches list；

Mixing Prefetching Model generation module, for based on user's history behavioural information, calculate popular video prefetch list weights and Long-tail video prefetches list weights, and weighted linear fusion hot topic video prefetches list and long-tail video prefetches list, generates individual character Change mixing Prefetching Model；

Video preprocessor delivery block, for, in the case of ensureing user's currently viewing video smoothness, based on personalized mixed model, being User prefetches and comes one or more videos the most forward.