CN206712985U - A kind of new network video push system - Google Patents
A kind of new network video push system Download PDFInfo
- Publication number
- CN206712985U CN206712985U CN201720494454.8U CN201720494454U CN206712985U CN 206712985 U CN206712985 U CN 206712985U CN 201720494454 U CN201720494454 U CN 201720494454U CN 206712985 U CN206712985 U CN 206712985U
- Authority
- CN
- China
- Prior art keywords
- hash
- text
- module
- video
- advertisement
- 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.)
- Expired - Fee Related
Links
Abstract
A kind of new network video push system is the utility model is related to, belongs to commending system and Hash learning art field.It is made up of across media hash modules, mixed type video recommendations subsystem, advertisement text extraction module, is characterized in：Across media hash modules are made up of image hash module, text hash module, across media relating modules, retrieval module；It is connected with mixed type video recommendations subsystem across the video inputs of media hash module, is connected across the text input of media hash module with advertisement text extraction module.By the possible advertisement hashed interested of beholder; realize that personalized advertisement pushes; it is selective and with strong points; attention rate of the beholder to push content is improved, strengthens attraction, extension gives more sustained attention the time; solves existing indifferenceization push specific aim, selective, interactive, relevance is poor; cause audient low to advertisement content attention rate, pay close attention to the problem of time is short, and pushing efficiency is low.
Description
Technical field
A kind of new network video push system is the utility model is related to, belongs to commending system and Hash learning art neck
Domain.
Background technology
Network video advertisement pushes the important component as transmission on Internet, changes traditional picture and text and propagates shape
Formula, it is a revolution of internet information.Network video advertisement push with break the time and space limitation, pass through culture every
Cut off from, the advantages of propagation resistance is small, scope is wide, various informative, beholder's interactivity is strong.At present, the master of network video advertisement push
Form is wanted to be pushed for forward type, i.e., this kind of advertisement pushing is intercutted before the video playback that beholder is clicked on, can efficiently reflected
Video transmission person wants the indices of tracking, such as clicking rate, per capita lasting viewing time, viewing cost etc..It is but this kind of
Advertisement takes unified dispensing mode mostly, there are problems that as follows：One is what is do not made full use of internet is interactive
And personalization, the ad content of owner's push is just as, beholder can not obtain oneself ad content interested,
So as to which the ad content attention rate to push is not high, influences beholder and make correct decisions.The second is push ad content with
Beholder's relevance interested is not strong, makes the advertisement of push relatively low by attention degree, is easily ignored or forgets.Thirdly
When being before multiple advertisements are placed in the video content of beholder oneself selection, the discharge order of advertisement is more random, sees
The person of seeing probably because the very first time do not reach oneself advertisement interested, and lose viewing ad content patience.
The content of the invention
The purpose of this utility model is, for above-mentioned the deficiencies in the prior art, there is provided one kind push content for a purpose is strong,
Attraction is big, and beholder gives more sustained attention time length, and propagation efficiency is high, is greatly improved concern of the beholder to advertisement content
The new network video push system of degree.
The utility model realizes above-mentioned purpose by following technical solution：
The new network video push system is by across media hash modules, mixed type video recommendations subsystem, advertisement text
Extraction module is formed, it is characterised in that：Across media hash modules are by image hash module, text hash module, across media associations
Module, retrieval module composition；It is connected across the video inputs of media hash module with mixed type video recommendations subsystem, across media
The text input of hash module is connected with advertisement text extraction module；
It is connected across an input of media relating module with the output end of image hash module, across media relating module
Another input is connected with the output end of text hash module；The input of image hash module and mixed type video recommendations point
The output end connection of system, the input of text hash module are connected with advertisement text extraction module；The inspection of image hash module
Rope signal output part, the recall signal output end of text hash module are respectively each connected with retrieval module；Retrieve module
Output end is connected with display terminal.
Described across media relating modules are made up of training picture library, training text storehouse, semantic association system, training picture library and
Training text storehouse stores training image collection and training text collection respectively, and semantic association system is by figure topic model and text subject mould
Type carries out semantic association, to carry out Hash study, makes training text and image while hashed, obtains figure hash function and text
Hash function；The module realizes the mutual retrieval between image and text, while makes image and text can be semantic complementary, improves inspection
Rope and push effect.
Described image hash module is made up of video receiver and figure Hash calculation unit, and video receiver, which obtains, to be recommended
Video content, figure Hash calculation unit calculate the Hash codes of video using figure hash function.The module receives mixed type video and pushed away
The video data of subsystem input is recommended, by the possible video hashed interested of beholder, and passes to retrieval module.
Described text hash module is made up of memory unit and text Hash calculation unit, and memory unit obtains wide
Content of text is accused, text Hash calculation unit calculates the Hash codes of text using text hash function.The module receives advertisement text
The text message of this extraction module input, by content of text hashed, passes to retrieval module.
Described retrieval module is made up of figure Hash receiver, text Hash receiver and Hash comparison unit, figure Hash
Receiver obtains the figure Hash codes of video, and text Hash receiver obtains this paper Hash codes of advertisement, and Hash comparison unit calculates
The Hamming distance of two Hash codes of input.The module can based on input video Hash arrange, calculate and the Hash row Hamming distance
Text Hash from minimum arranges.
The beneficial effect of the utility model compared with prior art is：
The new network video push system, by towards higher-dimension across media data hash algorithm, by beholder may
The content of text hashed of video ads interested, the personalized advertisement push for being different from prior art is realized, pushes content
It is with strong points, attention rate of the beholder to advertisement is substantially increased, strengthens attraction, when extension beholder gives more sustained attention
Between, propagation efficiency is effectively improved, expands propagation face and propagating influence.Solve existing indifferenceization push specific aim,
It is selective, interactive, relevance is poor, cause audient to push content attention rate it is low, concern the time it is short, pushing efficiency is low to ask
Topic.
Brief description of the drawings
Fig. 1 is a kind of overall structure diagram of new network video push system.
In figure：1st, across media hash modules, 2, mixed type video recommendations subsystem, 3, advertisement text extraction module, 101,
Across media relating modules, 102, image hash module, 103, text hash module, 104, retrieval module.
Embodiment
Specific embodiment of the present utility model is described in further detail below in conjunction with the accompanying drawings：
The new network video push system is by across media hash modules 1, mixed type video recommendations subsystem 2, advertisement text
This extraction module 3 is formed, and across media hash modules 1 are by across media relating modules 101, image hash module 102, text Hash mould
Block 103, retrieval module 104 form；The video inputs of across media hash modules 1 are connected with mixed type video recommendations subsystem 2,
The text input of across media hash modules 1 is connected with advertisement text extraction module 3；
One input of across media relating modules 101 is connected with the output end of image hash module 102, across media associations
Another input of module 101 is connected with the output end of text hash module 103；The input of image hash module 102 with
The output end connection of mixed type video recommendations subsystem 2, input and the advertisement text extraction module 3 of text hash module 103
Connection；The recall signal output end of image hash module 102, the recall signal output end of text hash module 103 are respectively each
It is connected from retrieval module 104；The output end of retrieval module 104 is connected with display terminal.
Described across media relating modules 101 are made up of training picture library, training text storehouse, semantic association system, training figure
Storehouse and training text storehouse store training image collection and training text collection respectively, and semantic association system is by figure topic model and text master
Inscribe model and carry out semantic association, to carry out Hash study, make training text and image hashed simultaneously, obtain figure hash function and
Text hash function；The module realizes the mutual retrieval between image and text, while makes image and text can be semantic complementary, carries
Height retrieval and push effect.
Described image hash module 102 is made up of video receiver and figure Hash calculation unit, and video receiver obtains
Recommend video content, figure Hash calculation unit calculates the Hash codes of video using figure hash function；The module receives mixed type and regarded
Frequency recommends the video data that subsystem 2 inputs, and by the possible video hashed interested of beholder, and passes to retrieval module
104。
Described text hash module 103 is made up of memory unit and text Hash calculation unit, and memory unit obtains
Advertisement text content is taken, text Hash calculation unit calculates the Hash codes of text using text hash function；The module receives wide
The text message of the input of Text Feature Extraction module 3 is accused, by advertisement text content Hash, passes to retrieval module 104.
Described retrieval module 104 is made up of figure Hash receiver, text Hash receiver and Hash comparison unit, and figure is breathed out
Uncommon receiver obtains the figure Hash codes of video, and text Hash receiver obtains this paper Hash codes of advertisement, Hash comparison unit meter
Calculate the Hamming distance of two Hash codes of input；The retrieval module 104 can based on input video Hash arrange, calculate and the Hash
The minimum text Hash row of row Hamming distance（Referring to Fig. 1）.
Hash study is different from traditional Hash, traditional hash algorithm be use random Harsh function by data be mapped as compared with
The binary value of long regular length, and Hash study is that data are mapped into binary string by mechanism of Machine Learning, can be significantly
Storage and the communication overhead of data are reduced, is from data so as to effectively improve efficiency Hash the destination of study of learning system
In learn hash function automatically, and then acquire the binary system Hash representation of data so that Hash codes keep former as much as possible
Neighbor relationships in space, specifically, each data point can be mapped to a compact binary coding, in former space
In similar 2 points can be mapped to 2 points similar in Hash code space.But, Hash study needs substantial amounts of training number
It is trained according to it, the accuracy of its coding could be improved.
The operation principle of the new network video push system is as described below：
Mixed type video recommendations subsystem 2 is used to read beholder's daily record, such as：The viewing record of beholder, beholder
Scoring to video etc..Mixed type video recommendations subsystem 2 is the mixed type video based on topic model and collaborative filtering
Commending system, it is made up of two modules, i.e. content filtering module and collaborative filtering module.Mixed type video recommendations subsystem 2
The basic procedure of proposed algorithm be：The historical data that beholder is watched to video first is pre-processed, and extracts beholder
Theme vector interested and characteristic vector, the recommending module of Cempetency-based education is established by information filtering pushing module, then
According to data such as beholder's interest characteristics, beholder's score data and currently viewing videos, built by collaborative filtering recommending module
Be based on the recommending module of collaborative filtering, extracts the arest neighbors of beholder and the arest neighbors of current accessed video, then integrates two
Individual recommending module is weighted summation operation, and Similarity Measure, generation are carried out with mixing recommended models to the video in video library
Top-N recommends video sequence.
Advertisement text extraction module 3 is used to extract the text message in input video advertisement, advertisement text extraction module 3
Extraction text message includes three phases：
1）, the method based on region segmentation finds the position of text in the picture in video frame images, to image carry out
String localization.
2）, being classified based on color characteristic by text image is converted to the binary picture that text pixel is 1, background pixel is 0
Picture, realize text segmentation.
3）, the text image of binaryzation is converted to the text message that computer can identify.
In the new network video push system, across the media hash modules 1 are specifically by across media relating modules
101st, image hash module 102, text hash module 103, retrieval module 104 form.
Wherein, across media relating modules 101 are connected with image hash module 102 and text hash module 103, by that will instruct
Practice the view data in data set and text data carries out feature extraction, image subject model and text subject can be respectively obtained
Model, a multimedia topic model then is obtained using the Semantic Connection of the two, then carry out Factorization, and carry out Hash
Study, make image Hash and text Hash be while mapped to same Hamming space, finally give corresponding figure hash function with
Text hash function.Enable image and text semantic complementary, improve the effect of retrieval and push.
Image hash module 102 and text hash module 103 have a figure hash function and a text Hash letter respectively
Number, image hash module 102 will recommend video transition to pass through text into Hash codes, text hash module 103 by figure hash function
Advertisement text is become Hash codes by this hash function.
Module 104 is retrieved for the figure Hash codes according to input, is found and the immediate text Hash codes of the Hash codes；It is logical
Cross and calculate text Hash codes with the Hamming distance between figure Hash codes, obtain the minimum several text Hash codes of distance, and by it
Arranged from small to large according to Hamming distance value, finally in order output corresponding to ad content.
The new network video push system, the content of text hashed of video ads that may be interested by beholder,
The personalized advertisement push for being different from prior art is realized, substantially increases degree of concern of the beholder to ad content.Push
Content for a purpose is strong, enhances the attraction to beholder, beholder is given more sustained attention time length, effectively increases advertizing
Efficiency.
Simply preferred embodiment of the present utility model described above, the example above illustrate not in essence of the present utility model
Appearance makees any formal limitation, and person of an ordinary skill in the technical field is new according to this practicality after this specification has been read
Any simple modification or deformation that the technical spirit of type is made to above embodiment, and possibly also with the disclosure above
Technology contents are changed or are modified to the equivalent embodiment of equivalent variations, still fall within the scope of technical solutions of the utility model
It is interior, without departing from spirit and scope of the present utility model.
Claims (5)
1. a kind of new network video push system, it is by across media hash modules（1）, mixed type video recommendations subsystem（2）、
Advertisement text extraction module（3）Form, it is characterised in that：Across media hash modules（1）By image hash module（102）, text
Hash module（103）, across media relating modules（101）, retrieval module（104）Composition；Across media hash modules（1）Video it is defeated
Enter end and mixed type video recommendations subsystem（2）Connection, across media hash modules（1）Text input and advertisement text extract
Module（3）Connection；
Across media relating modules（101）An input and image hash module（102）Output end connection, across media associations
Module（101）Another input and text hash module（103）Output end connection；Image hash module（102）It is defeated
Enter end and mixed type video recommendations subsystem（2）Output end connection, text hash module（103）Input and advertisement text
Extraction module（3）Connection；Image hash module（102）Recall signal output end, text hash module（103）Recall signal
Output end is respectively each with retrieving module（104）Connection；Retrieve module（104）Output end be connected with display terminal.
A kind of 2. new network video push system according to claim 1, it is characterised in that：Described across media associations
Module（101）It is made up of training picture library, training text storehouse, semantic association system；Training picture library and training text storehouse store respectively
Figure topic model and text subject model are carried out semantic association by training image collection and training text collection, semantic association system, with
Hash study is carried out, makes training text and image while hashed, obtains figure hash function and text hash function；Realize image
Mutual retrieval between text, while make image and text can be semantic complementary, improve retrieval and push effect.
A kind of 3. new network video push system according to claim 1, it is characterised in that：Described image Hash mould
Block（102）It is made up of video receiver and figure Hash calculation unit, video receiver, which obtains, recommends video content, figure Hash calculation
Unit calculates the Hash codes of video using figure hash function；To receive mixed type video recommendations subsystem（2）The video counts of input
According to by beholder's video hashed interested, and passing to retrieval module（104）.
A kind of 4. new network video push system according to claim 1, it is characterised in that：Described text Hash mould
Block（103）It is made up of memory unit and text Hash calculation unit, memory unit obtains advertisement text content, text Hash
Computing unit calculates the Hash codes of text using text hash function；To receive advertisement text extraction module（3）The text of input
Information, by advertisement text content Hash, pass to retrieval module（104）.
A kind of 5. new network video push system according to claim 1, it is characterised in that：Described retrieval module
（104）It is made up of figure Hash receiver, text Hash receiver and Hash comparison unit；Figure Hash receiver obtains the figure of video
Hash codes, text Hash receiver obtain this paper Hash codes of advertisement, and two Hash of input are calculated by Hash comparison unit
The Hamming distance of code, the video Hash row based on input, are calculated and the text Hash of Hash row Hamming distance minimum arranges.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201720494454.8U CN206712985U (en) | 2017-05-05 | 2017-05-05 | A kind of new network video push system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201720494454.8U CN206712985U (en) | 2017-05-05 | 2017-05-05 | A kind of new network video push system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN206712985U true CN206712985U (en) | 2017-12-05 |
Family
ID=60466419
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201720494454.8U Expired - Fee Related CN206712985U (en) | 2017-05-05 | 2017-05-05 | A kind of new network video push system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN206712985U (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114780789A (en) * | 2022-06-22 | 2022-07-22 | 山东建筑大学 | Assembly type component construction monitoring video positioning method based on natural language query |
-
2017
- 2017-05-05 CN CN201720494454.8U patent/CN206712985U/en not_active Expired - Fee Related
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114780789A (en) * | 2022-06-22 | 2022-07-22 | 山东建筑大学 | Assembly type component construction monitoring video positioning method based on natural language query |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9792620B2 (en) | System and method for brand monitoring and trend analysis based on deep-content-classification | |
| CN107852252B (en) | Fingerprint matching is recognized by automated content to change to detect channel | |
| CN101489139B (en) | Video advertisement correlation method and system based on visual saliency | |
| CN107862553A (en) | Advertisement real-time recommendation method, device, terminal device and storage medium | |
| US20170105051A1 (en) | Method and Apparatus for Increasing User Engagement with Video Advertisements and Content by Summarization | |
| TW201740295A (en) | Method for determining user behaviour preference, and method and device for presenting recommendation information | |
| CN102884538A (en) | Enriching online videos by content detection, searching, and information aggregation | |
| CN103365936A (en) | Video recommendation system and method thereof | |
| CN106547908A (en) | A kind of information-pushing method and system | |
| CN102332017A (en) | Method and device for displaying recommended information based on operation information in mobile device | |
| CN108174265B (en) | A kind of playback method, the apparatus and system of 360 degree of panoramic videos | |
| CN103942337A (en) | Video search system based on image recognition and matching | |
| CN205665733U (en) | Because cell neural network or convolution neural network's user identification system in advertising media input terminal | |
| CN105045818A (en) | Picture recommending method, apparatus and system | |
| CN105657558B (en) | A kind of information providing method, methods of exhibiting, device and smart television | |
| CN109783539A (en) | Usage mining and its model building method, device and computer equipment | |
| CN101789111A (en) | Site advertisement and system for real-time interaction of electronic advertisement data | |
| CN106570020A (en) | Method and apparatus used for providing recommended information | |
| CN106354859A (en) | Scenic spot recommendation device, method and system | |
| CN206712985U (en) | A kind of new network video push system | |
| CN100555283C (en) | A kind of directly at the dissemination method and the system of user's relevant information | |
| CN109612465A (en) | A kind of method, apparatus and its application creating personal scene various dimensions characteristic spectrum | |
| CN106709742A (en) | Data processing method and device of native advertisement | |
| CN110827063A (en) | Multi-strategy fused commodity recommendation method, device, terminal and storage medium | |
| CN102523512A (en) | Video output method with operable implicit content |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20171205 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |