CN117171390B

CN117171390B - Information pushing method and system based on big data

Info

Publication number: CN117171390B
Application number: CN202311190613.1A
Authority: CN
Inventors: 沈弘悦
Original assignee: Shenzhen Ferromagnetic Digital Technology Co ltd
Current assignee: Shenzhen Ferromagnetic Digital Technology Co ltd
Priority date: 2023-09-15
Filing date: 2023-09-15
Publication date: 2024-04-26
Anticipated expiration: 2043-09-15
Also published as: CN117171390A

Abstract

The invention provides an information pushing method and system based on big data, and relates to the technical field of artificial intelligence. In the invention, the video data to be pushed is subjected to data splitting operation to form a video data fragment set; performing key information mining operation on the video data fragment set, and outputting a key information description vector set; splitting the key information description vector set to form a key information description vector sub-set; carrying out information enrichment operation on the key information description vector sub-set to form an enriched description vector sub-set; carrying out combination operation on the enriched description vector sub-sets and outputting the enriched description vector sets; and carrying out aggregation operation on the enriched description vector set to form an aggregated key information description vector, carrying out push evaluation operation on the aggregated key information description vector, and outputting push evaluation result data. Based on the above, the reliability of video data push can be improved to some extent.

Description

Information pushing method and system based on big data

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to an information pushing method and system based on big data.

Background

Video data is an important multimedia data, and has many applications in the internet, for example, the video data can be pushed to a user based on the requirement or preference of the user, that is, the video data that may be required or interested by the user is sent to the user. However, in the prior art, during the process of pushing video data, the video data is generally analyzed to determine a corresponding pushing direction, and then, the corresponding pushing is performed based on the pushing direction, but the problem of low reliability of the determination of the pushing direction is existed, so that the problem of low reliability of pushing is easy to occur.

Disclosure of Invention

In view of the above, the present invention is directed to providing a method and a system for pushing information based on big data, so as to improve the reliability of video data pushing to a certain extent.

In order to achieve the above purpose, the embodiment of the present invention adopts the following technical scheme:

An information pushing method based on big data comprises the following steps:

Respectively carrying out data splitting operation on video data to be pushed based on each of a plurality of preset dividing lengths to form a video data fragment set corresponding to each of the dividing lengths, wherein for each dividing length, the video data fragment set corresponding to the dividing length comprises a plurality of video data fragments belonging to the dividing length, and each video data fragment comprises at least one frame of video frame to be pushed;

Performing key information mining operation on a plurality of video data fragments in a video data fragment set corresponding to each division length respectively to output a key information description vector set corresponding to each division length, wherein for each division length, the key information description vector set corresponding to the division length comprises key information description vectors mined by the plurality of video data fragments belonging to the division length;

For each division length, splitting a key information description vector set corresponding to the division length to form a plurality of key information description vector sub-sets corresponding to the division length, wherein each key information description vector sub-set comprises at least one original key information description vector, and each original key information description vector corresponds to one video data segment;

performing information enrichment operation on original key information description vectors in the plurality of key information description vector sub-sets to form an enriched description vector sub-set corresponding to the plurality of key information description vector sub-sets, wherein the enriched description vector sub-set comprises at least one enriched key information description vector, and each enriched key information description vector corresponds to one original key information description vector in the key information description vector sub-set corresponding to the enriched description vector sub-set;

Based on the precedence relation among the plurality of key information description vector sub-sets, carrying out combination operation on the formed plurality of enriched description vector sub-sets so as to output an enriched description vector set corresponding to the segmentation length, wherein the video data fragment set, the key information description vector sub-set, the enriched description vector sub-set and the enriched description vector set all belong to an ordered set of description vectors;

Performing aggregation operation on the rich description vector set corresponding to each of the plurality of division lengths to form an aggregation key information description vector corresponding to the video data to be pushed, and performing push evaluation operation on the aggregation key information description vector to output push evaluation result data corresponding to the video data to be pushed, wherein the push evaluation result data is used for reflecting a push direction of performing data push operation on the video data to be pushed.

In some preferred embodiments, in the big data based information pushing method, the step of performing an information enrichment operation on the original key information description vectors in the plurality of key information description vector sub-sets to form enriched description vector sub-sets corresponding to the plurality of key information description vector sub-sets includes:

aiming at any one of the key information description vector sub-sets, respectively carrying out information enrichment operation on each original key information description vector based on a plurality of original key information description vectors in the key information description vector sub-set so as to output an intermediate key information description vector corresponding to each original key information description vector;

Combining the plurality of output intermediate key information description vectors to form an intermediate description vector sub-set corresponding to the key information description vector sub-set;

and based on the importance characterization parameters corresponding to the segmentation lengths, adjusting the intermediate description vector sub-sets to form a rich description vector sub-set corresponding to the key information description vector sub-set.

determining the distribution coordinates of the subsets corresponding to the key information description vector subsets;

And carrying out data integration operation on each key information description vector sub-set and corresponding sub-set distribution coordinates so as to output a rich description vector sub-set corresponding to each key information description vector sub-set, wherein the sub-set distribution coordinates are used for reflecting the distribution coordinates of the corresponding key information description vector sub-set in the plurality of key information description vector sub-sets.

In some preferred embodiments, in the big data based information pushing method, the step of performing a data integration operation on the distribution coordinates of each of the key information description vector subsets and the corresponding subset to output a rich description vector subset corresponding to each of the key information description vector subsets includes:

Performing vector space conversion operation on each original key information description vector in any one of the key information description vector sub-sets to form a converted key information description vector corresponding to each original key information description vector, wherein the vector space conversion operation at least comprises filtering operation and linear mapping operation;

Vector combination operation is carried out on the basis of the formed conversion key information description vector so as to form a conversion description vector sub-set corresponding to the key information description vector sub-set;

And carrying out data integration operation on the transformation description vector sub-sets corresponding to the key information description vector sub-sets and the subset distribution coordinates corresponding to the key information description vector sub-sets so as to output the enrichment description vector sub-sets corresponding to the key information description vector sub-sets.

In some preferred embodiments, in the big data based information pushing method, each of the key information description vector subsets includes a plurality of original key information description vectors;

The step of performing information enrichment operation on the original key information description vectors in the plurality of key information description vector sub-sets to form enriched description vector sub-sets corresponding to the plurality of key information description vector sub-sets includes:

Respectively carrying out data integration operation on each key information description vector sub-set and the distribution coordinates of the corresponding sub-set so as to form an integrated description vector sub-set corresponding to each key information description vector sub-set, wherein each integrated description vector sub-set comprises a plurality of integrated key information description vectors, and each integrated key information description vector corresponds to one original key information description vector in the corresponding key information description vector sub-set;

Vector combination operation is carried out on a plurality of integrated key information description vectors on the same distribution coordinates in each formed plurality of integrated description vector sub-sets so as to form a corresponding combined description vector sub-set, and a plurality of corresponding combined description vector sub-sets are output;

Performing adjustment operation on each combined description vector sub-set based on the corresponding importance reflection parameters respectively to form an adjustment description vector sub-set corresponding to each combined description vector sub-set, wherein the adjustment description vector sub-set comprises a plurality of adjustment key information description vectors, and each adjustment key information description vector corresponds to one integration key information description vector in the corresponding combined description vector sub-set;

Based on the formed adjustment key information description vectors on the same distribution coordinates of each of the plurality of adjustment description vector sub-sets, combining to form a corresponding one of the enriched description vector sub-sets to form a corresponding plurality of enriched description vector sub-sets.

In some preferred embodiments, in the big data based information pushing method, after the step of performing a vector combination operation on the plurality of integrated key information description vectors on each of the same distribution coordinates in the formed plurality of integrated description vector subsets to form a corresponding one of the combined description vector subsets to output a corresponding plurality of combined description vector subsets, the big data based information pushing method further includes:

Analyzing vector matching characterization parameters among a plurality of integrated key information description vectors in each combination description vector sub-set;

And respectively determining importance reflecting parameters corresponding to each combined description vector sub-set according to vector matching characteristic parameters among a plurality of integrated key information description vectors in each combined description vector sub-set.

In some preferred embodiments, in the big data based information pushing method, each of the key information description vector subsets includes a plurality of original key information description vectors, and any two key information description vector subsets having an adjacency relationship in the plurality of key information description vector subsets include at least one identical original key information description vector;

the step of performing a combination operation on the formed multiple enriched description vector subsets based on the precedence relation among the multiple key information description vector subsets to output the enriched description vector sets corresponding to the segmentation length includes:

Based on the precedence relation among the plurality of key information description vector sub-sets, carrying out aggregation operation on the rich key information description vectors corresponding to the same original key information description vector in every two rich description vector sub-sets with adjacent relation;

And combining the key information description vectors and other key information description vectors after the plurality of the enrichment description vector subsets are aggregated to form an enrichment description vector set corresponding to the segmentation length.

In some preferred embodiments, in the foregoing big data based information pushing method, the step of performing an aggregation operation on the rich key information description vectors corresponding to the same original key information description vector in each two rich description vector sub-sets having an adjacency relationship based on the precedence relationship between the plurality of key information description vector sub-sets includes:

performing aggregation operation on the rich key information description vector and the corresponding original key information description vector in each rich description vector sub-set to form a corresponding aggregation depth description vector;

and based on the precedence relation among the plurality of key information description vector sub-sets, carrying out aggregation operation on the aggregation depth description vectors corresponding to the same original key information description vector in each two rich description vector sub-sets with the adjacency relation.

In some preferred embodiments, in the big data based information pushing method, the enriched description vector set and the aggregated key information description vector each include a row vector parameter and a column vector parameter;

the step of performing an aggregation operation on the enriched description vector set corresponding to each of the plurality of partition lengths to form an aggregated key information description vector corresponding to the video data to be pushed, and performing a push evaluation operation on the aggregated key information description vector to output push evaluation result data corresponding to the video data to be pushed, where the push evaluation result data is used for reflecting a push direction of performing a data push operation on the video data to be pushed, includes:

Marking a rich description vector set corresponding to a specified segmentation length in the plurality of segmentation lengths to be marked as the specified rich description vector set;

adjusting the enriched description vector sets corresponding to the other division lengths except the appointed division length in the plurality of division lengths to form other enriched description vector sets with the same vector size as the appointed enriched description vector set;

Performing cascading combination operation on the appointed enrichment description vector set and the other enrichment description vector sets to form an aggregation key information description vector corresponding to the video data to be pushed;

Performing pushing evaluation operation on the aggregated key information description vector to output pushing evaluation result data corresponding to the video data to be pushed, wherein the pushing evaluation result data is used for reflecting the pushing direction of performing data pushing operation on the video data to be pushed;

The step of performing a cascade combination operation on the specified rich description vector set and the other rich description vector sets to form an aggregate key information description vector corresponding to the video data to be pushed includes:

And performing cascading combination operation on row vector parameters or column vector parameters with identical vector distribution coordinates in the appointed enrichment description vector set and the other enrichment description vector sets to form a plurality of cascading combination vector parameter sets, and combining to form an aggregation key information description vector corresponding to the video data to be pushed based on the cascading combination vector parameter sets.

The embodiment of the invention also provides an information pushing system based on big data, which comprises a processor and a memory, wherein the memory is used for storing a computer program, and the processor is used for executing the computer program so as to realize the information pushing method based on big data.

According to the information pushing method and system based on big data, the video data to be pushed can be subjected to data splitting operation to form a video data fragment set; performing key information mining operation on the video data fragment set, and outputting a key information description vector set; splitting the key information description vector set to form a key information description vector sub-set; carrying out information enrichment operation on the key information description vector sub-set to form an enriched description vector sub-set; carrying out combination operation on the enriched description vector sub-sets and outputting the enriched description vector sets; and carrying out aggregation operation on the enriched description vector set to form an aggregated key information description vector, carrying out push evaluation operation on the aggregated key information description vector, and outputting push evaluation result data. Based on the foregoing, since the aggregate key information description vector, which is the basis of the push evaluation operation, is the result of the information enrichment operation based on the video data segment set formed by performing the data splitting operation on the push video data with a plurality of division lengths, the aggregate key information description vector has stronger information characterization capability and higher reliability, so that the reliability of the determined video data push direction can be improved to a certain extent, the problem of low reliability of the determined video data push direction in the prior art is improved, the reliability of the video data push is improved to a certain extent, and the problem of poor reliability of the video data push in the prior art is improved.

In order to make the above objects, features and advantages of the present invention more comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

Fig. 1 is a block diagram of a big data based information pushing system according to an embodiment of the present invention.

Fig. 2 is a flowchart illustrating steps included in the big data based information pushing method according to the embodiment of the present invention.

Fig. 3 is a schematic diagram of each module included in the big data based information pushing device according to the embodiment of the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, but not all embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.

Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

As shown in fig. 1, the embodiment of the invention provides an information pushing system based on big data. Wherein the big data based information push system may comprise a memory and a processor.

In detail, the memory and the processor are electrically connected directly or indirectly to realize transmission or interaction of data. For example, electrical connection may be made to each other via one or more communication buses or signal lines. The memory may store at least one software functional module (computer program) that may exist in the form of software or firmware. The processor may be configured to execute the executable computer program stored in the memory, so as to implement the big data based information pushing method provided by the embodiment of the present invention.

It should be appreciated that in some possible embodiments, the Memory may be, but is not limited to, random access Memory (Random Access Memory, RAM), read Only Memory (ROM), programmable Read Only Memory (Programmable Read-Only Memory, PROM), erasable Read Only Memory (Erasable Programmable Read-Only Memory, EPROM), electrically erasable Read Only Memory (Electric Erasable Programmable Read-Only Memory, EEPROM), and the like.

It should be appreciated that in some possible embodiments, the processor may be a general purpose processor including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), a System on Chip (SoC), etc.; but also Digital Signal Processors (DSPs), application Specific Integrated Circuits (ASICs), field Programmable Gate Arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

It should be appreciated that in some possible embodiments, the big data based information push system may be a server with data processing capabilities.

With reference to fig. 2, the embodiment of the invention further provides an information pushing method based on big data, which can be applied to the information pushing system based on big data. The method steps defined by the flow related to the big data based information pushing method can be realized by the big data based information pushing system. The specific flow shown in fig. 2 will be described in detail.

Step S110, based on each of a plurality of predetermined dividing lengths, respectively, the video data to be pushed is subjected to data splitting operation to form a video data segment set corresponding to each of the plurality of dividing lengths.

In the embodiment of the present invention, the big data based information pushing system may perform a data splitting operation on the video data to be pushed based on each of a plurality of predetermined splitting lengths, so as to form a video data segment set corresponding to each of the plurality of splitting lengths. For each of the segment lengths, the set of video data segments corresponding to the segment length includes a plurality of video data segments belonging to the segment length, and a length (such as a duration or a number of included video frames to be pushed) of each of the plurality of video data segments is equal to the segment length, and each of the video data segments includes at least one frame of video frames to be pushed. The video data to be pushed may include a plurality of frames of video frames to be pushed. For example, based on a first division length, performing a data splitting operation on the video data to be pushed to form a video data segment set corresponding to the first division length, and based on a second division length, performing a data splitting operation on the video data to be pushed to form a video data segment set corresponding to the second division length. In addition, the specific numerical value of each division length is not limited, and can be configured according to actual requirements.

Step S120, performing a key information mining operation on a plurality of video data segments in the video data segment set corresponding to each segment length, so as to output a key information description vector set corresponding to each segment length.

In the embodiment of the invention, the big data based information pushing system can perform key information mining operation on a plurality of video data fragments in the video data fragment set corresponding to each division length respectively so as to output a key information description vector set corresponding to each division length. For each of the division lengths, the key information description vector set corresponding to the division length includes key information description vectors mined by a plurality of video data segments belonging to the division length. The video data segments may be subjected to key information mining operations, such as mapping and filtering operations of feature spaces, to obtain corresponding key information description vectors, and then, combined and ordered according to a precedence relationship between the video data segments, so as to form corresponding key information description vector sets. In addition, for different segmentation lengths, key information mining operations can be performed on a plurality of video data segments in the corresponding video data segment set based on different key information mining units, respectively. For example, the key information mining unit 1 performs the key information mining operation on each of the plurality of video data segments in the set of frequency data segments corresponding to the division length 1, the key information mining unit 2 performs the key information mining operation on each of the plurality of video data segments in the set of frequency data segments corresponding to the division length 2, and the key information mining unit 3 performs the key information mining operation on each of the plurality of video data segments in the set of frequency data segments corresponding to the division length 3.

Step S130, for each of the division lengths, splitting the set of key information description vectors corresponding to the division length to form a plurality of sets of key information description vectors corresponding to the division length.

In the embodiment of the present invention, the big data based information push system may split, for each of the partition lengths, a set of key information description vectors corresponding to the partition length, so as to form a plurality of sets of key information description vectors corresponding to the partition length. Each keyinformation description vector subset comprises at least one original keyinformation description vector, each of said original keyinformation description vectors corresponding to one of said video data segments. For example, for a plurality of key information description vector sub-sets corresponding to any one of the division lengths, each key information description vector sub-set includes a plurality of original key information description vectors, and any two key information description vector sub-sets having an adjacency relationship among the plurality of key information description vector sub-sets include at least one identical original key information description vector. In addition, in the plurality of key information description vector sub-sets, the number of original key information description vectors included in each key information description vector sub-set may be the same, and the number of the same original key information description vectors in any two key information description vector sub-sets having an adjacency relationship may be equal to 0.1, 0.2, 0.3, 0.4 times, or the like of the number of the original key information description vectors included in the key information description vector sub-sets. For example, for a key information description vector 1 and a key information description vector sub-set 2 having an adjacency relationship in a plurality of key information description vector sub-sets corresponding to any one of the division lengths, the key information description vector sub-set 1 and the key information description vector sub-set 2 each include five original key information description vectors, the key information description vector sub-set 1 includes an original key information description vector 1, an original key information description vector 2, an original key information description vector 3, an original key information description vector 4, an original key information description vector 5, and the key information description vector sub-set 2 includes an original key information description vector 3, an original key information description vector 4, an original key information description vector 5, an original key information description vector 6, and an original key information description vector 7, that is, the key information description vector sub-set 1 and the key information description vector sub-set 2 include 3 identical original key information description vectors.

And step S140, carrying out information enrichment operation on the original key information description vectors in the plurality of key information description vector sub-sets to form enriched description vector sub-sets corresponding to the plurality of key information description vector sub-sets.

In the embodiment of the invention, the big data-based information pushing system can perform information enrichment operation on the original key information description vectors in the plurality of key information description vector sub-sets to form enriched description vector sub-sets corresponding to the plurality of key information description vector sub-sets. The subset of rich description vectors comprises at least one rich key information description vector, and each rich key information description vector corresponds to one original key information description vector in the subset of key information description vectors corresponding to the subset of rich description vectors. That is, the information of the original key information description vector can be enhanced, so that the formed rich description vector sub-set has stronger characterization capability and is more reliable.

And step S150, based on the precedence relation among the plurality of key information description vector sub-sets, carrying out combination operation on the plurality of formed rich description vector sub-sets so as to output the rich description vector sets corresponding to the segmentation length.

In the embodiment of the invention, the big data-based information pushing system can perform a combination operation on the formed multiple enriched description vector sub-sets based on the precedence relationship among the multiple key information description vector sub-sets so as to output the enriched description vector set corresponding to the segmentation length. For example, the key information description vector sub-set 1 corresponds to the rich description vector sub-set 1, the key information description vector sub-set 2 corresponds to the rich description vector sub-set 2, and the key information description vector sub-set 3 corresponds to the rich description vector sub-set 3, so that in the case of having a precedence relationship of the key information description vector sub-set 1, the key information description vector sub-set 2, and the key information description vector sub-set 3, there is a precedence relationship of the rich description vector sub-set 1, the rich description vector sub-set 2, and the rich description vector sub-set 3 in the rich description vector sub-set. The video data segment set, the key information description vector sub-set, the rich description vector sub-set and the rich description vector set all belong to an ordered set of description vectors, namely, are ordered according to a precedence relationship.

Step S160, performing an aggregation operation on the enriched description vector set corresponding to each of the plurality of partition lengths to form an aggregate key information description vector corresponding to the video data to be pushed, and performing a push evaluation operation on the aggregate key information description vector to output push evaluation result data corresponding to the video data to be pushed.

In the embodiment of the present invention, the big data based information pushing system may perform an aggregation operation on the enriched description vector set corresponding to each of the multiple partition lengths to form an aggregate key information description vector corresponding to the video data to be pushed, and perform a pushing evaluation operation on the aggregate key information description vector to output pushing evaluation result data corresponding to the video data to be pushed. The pushing evaluation result data is used for reflecting the pushing direction of the data pushing operation of the video data to be pushed. For example, the aggregate key information description vector may be processed, such as a full connection process and an excitation mapping process, based on a push evaluation unit included in the neural network, so as to obtain corresponding push evaluation result data, where the excitation mapping process may be implemented by a function such as softmax. In addition, the push evaluation result data may be used to reflect the effectiveness degree of pushing the video data to be pushed to the target user, so when the push evaluation operation is performed, a user feature vector (such as a preference characterizing that the user has in a video data dimension) of the target user may be further combined with the aggregate key information description vector, and then, full connection processing and excitation mapping processing are performed on the vector obtained by the combination. The user feature vector can be obtained by mining key information based on user identity information of the target user and historical browsing videos.

Based on the foregoing (i.e., the foregoing step S110-step S160), since the aggregate key information description vector, which is the basis of the push evaluation operation, is the result of the information enrichment operation based on the video data segment set formed by performing the data splitting operation on the push video data with multiple splitting lengths, the information characterizing capability of the aggregate key information description vector is stronger and the reliability is higher, so that the reliability of the determined push direction of the video data can be improved to a certain extent, the problem that the reliability of the determined push direction of the video data in the prior art is not high is solved, so that the reliability of the push of the video data is improved to a certain extent, and the problem that the reliability of the push of the video data in the prior art is not good is solved.

It should be understood that, in some possible embodiments, step S120 in the foregoing description, that is, the step of performing the key information mining operation on the plurality of video data segments in the video data segment set corresponding to each of the division lengths to output the key information description vector set corresponding to each of the division lengths, further includes the following specific implementation matters:

For any video data segment, mapping operation of feature space is carried out on each video frame to be pushed in a plurality of video frames to be pushed included in the video data segment so as to form a plurality of video frame mapping vectors corresponding to the plurality of video frames to be pushed;

performing superposition operation or mean value operation on the plurality of video frame mapping vectors to form a first local key information description vector corresponding to the video data segment;

performing a pairwise combination operation on at least three video frame mapping vectors in the plurality of video frame mapping vectors to form a plurality of corresponding mapping vector combinations, wherein each two video frame mapping vectors included in the mapping vector combinations are not identical;

For each mapping vector combination, performing cross multiplication operation on two video frame mapping vectors included in the mapping vector combination to output a cross multiplication mapping vector corresponding to the mapping vector combination, and performing point multiplication operation on two video frame mapping vectors included in the mapping vector combination to output a mapping vector point multiplication operation parameter corresponding to the mapping vector combination;

performing weighted superposition operation on cross multiplication mapping vectors corresponding to each mapping vector combination based on the mapping vector point multiplication operation parameters corresponding to each mapping vector combination to form a second local key information description vector corresponding to the video data segment;

Performing cascading combination operation on the plurality of video frame mapping vectors to form cascading combination video frame mapping vectors corresponding to the plurality of video frame mapping vectors, performing full connection operation on the cascading combination video frame mapping vectors to form corresponding full connection video frame mapping vectors, and performing nonlinear activation operation on the full connection video frame mapping vectors to form third local key information description vectors corresponding to the video data segments;

And performing superposition operation or mean operation on the first local key information description vector, the second local key information description vector and the third local key information description vector to form a key information description vector corresponding to the video data segment, wherein when the vector sizes of the first local key information description vector, the second local key information description vector and the third local key information description vector are different, the vector sizes can be adjusted first, so that the vector sizes of the adjusted first local key information description vector, the adjusted second local key information description vector and the adjusted third local key information description vector are the same, then, performing superposition operation or mean operation to form a key information description vector corresponding to the video data segment, and based on the key information description vector corresponding to each video data segment can be obtained after each video data segment is processed as above, so that the key information description vector corresponding to each video data segment can be combined to form a key information segmentation description vector set corresponding to the length.

It should be understood that, in some possible embodiments, step S140 in the foregoing description, that is, the step of performing an information enrichment operation on the original key information description vectors in the plurality of key information description vector sub-sets to form enriched description vector sub-sets corresponding to the plurality of key information description vector sub-sets, may further include the following specific implementation matters:

for any one of the key information description vector sub-sets, based on a plurality of original key information description vectors in the key information description vector sub-set, respectively performing information enrichment operation on each original key information description vector to output an intermediate key information description vector corresponding to each original key information description vector, for example, for each original key information description vector, focus feature analysis operation can be performed on the aimed original key information description vector based on other original key information description vectors to form a corresponding intermediate key information description vector;

Combining the outputted plurality of intermediate key information description vectors to form an intermediate description vector sub-set corresponding to the subset of key information description vectors, that is, the intermediate description vector sub-set may include the plurality of intermediate key information description vectors;

and based on the importance characterization parameters corresponding to the segmentation length, performing an adjustment operation on the intermediate description vector sub-set to form a rich description vector sub-set corresponding to the key information description vector sub-set, wherein the importance characterization parameters can be configured according to the segmentation length, for example, the importance characterization parameters can be positively correlated with the segmentation length, for example, the intermediate description vector sub-set and the importance characterization parameters can be multiplied, and the importance characterization parameters can be the same as the intermediate description vector sub-set and comprise row parameters and column parameters (can be formed in a network optimization process of a corresponding neural network).

It should be understood that, in some possible embodiments, the step of performing, for any one of the key information description vector subsets, an information enrichment operation on each of the original key information description vectors based on a plurality of original key information description vectors in the key information description vector subset to output an intermediate key information description vector corresponding to each of the original key information description vectors may further include the following specific implementation matters:

Aiming at any one of the key information description vector sub-sets, taking one original key information description vector of a plurality of original key information description vectors in the key information description vector sub-set as a description vector to be processed, taking each other original key information description vector as an association description vector, and carrying out average calculation on each association description vector to output an average association description vector corresponding to the description vector to be processed;

And carrying out focusing characteristic analysis operation on the description vector to be processed based on the mean value associated description vector so as to form an intermediate key information description vector corresponding to the description vector to be processed.

And carrying out data integration operation on each key information description vector sub-set and corresponding sub-set distribution coordinates so as to output a rich description vector sub-set corresponding to each key information description vector sub-set, wherein the sub-set distribution coordinates are used for reflecting the distribution coordinates of the corresponding key information description vector sub-set in the plurality of key information description vector sub-sets, such as the first key information description vector sub-set, the second key information description vector sub-set and the like.

It should be understood that, in some possible embodiments, the step of performing a data integration operation on each of the key information description vector subsets and the distribution coordinates of the corresponding subset to output a corresponding rich description vector subset of each of the key information description vector subsets may further include the following specific implementation matters:

For any one of the key information description vector subsets, performing a vector space conversion operation on each original key information description vector in the key information description vector subset to form a converted key information description vector corresponding to each original key information description vector, where the vector space conversion operation includes at least a filtering operation and a linear mapping operation, for example, the original key information description vector is subjected to a filtering operation by performing a filtering matrix formed by network optimization, and then the obtained filtering description vector may be subjected to a linear mapping operation to form a corresponding converted key information description vector, where the linear mapping operation may include a weighting parameter and an offset parameter formed by a network optimization operation (the network optimization operation may be performed based on exemplary video data and a corresponding labeled pushing direction), and for example, the obtained result and the offset parameter may be added after multiplying the filtering description vector and the weighting parameter;

Vector combination operation is carried out based on the formed conversion key information description vectors so as to form a conversion description vector sub-set corresponding to the key information description vector sub-set, namely, the conversion description vector sub-set comprises all the conversion key information description vectors;

And performing data integration operation on the transformation description vector sub-set corresponding to the key information description vector sub-set and the subset distribution coordinate corresponding to the key information description vector sub-set to output the enrichment description vector sub-set corresponding to the key information description vector sub-set, wherein, for example, vectorization operation can be performed on the subset distribution coordinate first, then superposition operation can be performed on the vectorization operation result and the transformation description vector sub-set, and finally, full connection processing is performed on the superposition operation result to form the corresponding enrichment description vector sub-set.

It should be understood that, in some possible embodiments, each of the key information description vector subsets includes a plurality of original key information description vectors, based on which step S140 in the above description, that is, the step of performing an information enrichment operation on the original key information description vectors in the plurality of key information description vector subsets to form enriched description vector subsets corresponding to the plurality of key information description vector subsets, further includes the following specific implementation matters:

Respectively performing data integration operation on each key information description vector sub-set and a corresponding sub-set distribution coordinate to form an integrated description vector sub-set corresponding to each key information description vector sub-set, wherein each integrated description vector sub-set comprises a plurality of integrated key information description vectors, each integrated key information description vector corresponds to an original key information description vector in the corresponding key information description vector sub-set, the integrated key information description vectors can be formed by performing integration operation on the distribution coordinates of the original key information description vector and the video data fragments corresponding to the original key information description vector in the corresponding sub-set, and can be processed in a previous way or different, for example, the vectorization results of the distribution coordinates of the original key information description vector and the video data fragments corresponding to the original key information description vector in the corresponding sub-set can be subjected to cascade combination operation, and then the result of the cascade combination operation can be subjected to self-focusing feature analysis operation to obtain the integrated key information description vector;

Vector combining the plurality of integrated key information description vectors on each of the same distribution coordinates in the plurality of formed integrated description vector sub-sets to form a corresponding one of the plurality of combined description vector sub-sets to output a corresponding one of the plurality of combined description vector sub-sets, for example, the integrated key information description vector of the first distribution coordinate in the first integrated description vector sub-set, the integrated key information description vector of the first distribution coordinate in the second integrated description vector sub-set, the integrated key information description vector of the first distribution coordinate in the third integrated description vector sub-set, and the integrated key information description vector of the first distribution coordinate in the third integrated description vector sub-set may be combined to form a first combined description vector sub-set, and the integrated key information description vector of the second distribution coordinate in the first integrated description vector sub-set, the integrated key information description vector of the second distribution coordinate in the second integrated description vector sub-set may be combined to form a second combined description vector sub-set;

Based on the formed adjustment key information description vectors on the same distribution coordinates of each of the plurality of adjustment description vector sub-sets, a corresponding one of the enriched description vector sub-sets is formed by combining to form a corresponding plurality of enriched description vector sub-sets, as previously described in relation thereto.

It should be appreciated that, in some possible embodiments, after the step of performing a vector combining operation on the plurality of integrated key information description vectors formed on each of the plurality of integrated description vector sub-sets and on the same distribution coordinates to form a corresponding one of the plurality of combined description vector sub-sets to output the corresponding plurality of combined description vector sub-sets, the big data based information pushing method may further include the following specific implementation contents:

analyzing vector matching characteristic parameters, such as cosine similarity among vectors, among a plurality of integrated key information description vectors in each combination description vector sub-set;

According to the vector matching characteristic parameters among the plurality of integrated key information description vectors in each combined description vector sub-set, importance reflection parameters corresponding to each combined description vector sub-set are respectively determined, for example, for each combined description vector sub-set, mean value calculation can be carried out on the vector matching characteristic parameters among the plurality of integrated key information description vectors in the combined description vector sub-set, and the importance reflection parameters corresponding to the combined description vector sub-set are obtained.

It should be understood that, in some possible embodiments, each of the key information description vector sub-sets may include a plurality of original key information description vectors, where any two key information description vector sub-sets having an adjacency relationship in the plurality of key information description vector sub-sets include at least one identical original key information description vector, based on this, step S150 in the foregoing description, that is, the step of combining the formed plurality of enriched description vector sub-sets based on the precedence relationship between the plurality of key information description vector sub-sets to output the enriched description vector sets corresponding to the splitting lengths may further include the following specific implementation matters:

And performing a combination operation on the key information description vector aggregated by the plurality of enriched description vector subsets and other key information description vectors to form an enriched description vector set corresponding to the segmentation length, for example, a first enriched description vector set comprises an enriched key information description vector 1, an enriched key information description vector 2 and an enriched key information description vector 3, a second enriched description vector set comprises an enriched key information description vector 4, an enriched key information description vector 5 and an enriched key information description vector 6, wherein the enriched key information description vector 3 is identical (i.e. the same as the original key information description vector corresponding to the enriched key information description vector 4), so that the enriched key information description vector 3 and the enriched key information description vector 4 can be subjected to an aggregation operation to form an aggregated key information description vector 1, and based on the enriched description vector set, the formed enriched description vector set can comprise the enriched key information description vector 1, the enriched key information description vector 2, the aggregated key information description vector 1, the enriched key information description vector 5 and the enriched key information description vector 6.

It should be understood that, in some possible embodiments, the step of aggregating, based on the precedence relationship between the multiple key information description vector sub-sets, the rich key information description vectors of each two rich description vector sub-sets having the adjacency relationship, which correspond to the same original key information description vector, may further include the following specific implementation matters:

Performing aggregation operation on the rich key information description vector and the corresponding original key information description vector in each rich description vector sub-set to form a corresponding aggregation depth description vector, for example, performing superposition operation on the rich key information description vector and the corresponding original key information description vector to form an aggregation depth description vector;

Based on the precedence relation between the multiple key information description vector sub-sets, the aggregation depth description vectors of every two richness description vector sub-sets with the adjacency relation corresponding to the same original key information description vector are aggregated, for example, in the process of aggregating the richness key information description vector 3 and the richness key information description vector 4, the richness key information description vector 3 and the corresponding original key information description vector may be aggregated first to form an aggregation depth description vector 3, the richness key information description vector 4 and the corresponding original key information description vector are aggregated to form an aggregation depth description vector 4, then the aggregation depth description vector 3 and the aggregation depth description vector 4 may be subjected to superposition operation to form an aggregated key information description vector 1, or in other embodiments, the aggregation depth description vector 4 may also be subjected to focus feature analysis operation based on the aggregation depth description vector 3, then the focus feature analysis operation may be performed on the aggregation depth description vector 3, the results of the two focus feature analysis operations may be subjected to cascade combination operation, and the cascade combination operation may be performed on the results of the focus feature analysis operation, and the cascade operation may be performed on the aggregation depth description vector 4 to form a key information description vector 1.

It should be understood that, in some possible embodiments, the enriching description vector set and the aggregating key information description vector include row vector parameters and column vector parameters, that is, have a row direction and a column direction, and belong to two-dimensional vectors, based on this, step S160 in the foregoing description, that is, the aggregating operation is performed on the enriching description vector set corresponding to each of the plurality of partition lengths to form an aggregating key information description vector corresponding to the video data to be pushed, and the step of performing the pushing evaluation operation on the aggregating key information description vector to output pushing evaluation result data corresponding to the video data to be pushed may further include the following specific implementation contents:

Marking a rich description vector set corresponding to a designated partition length in the plurality of partition lengths to be marked as the designated rich description vector set, wherein the designated partition length may be any partition length in the plurality of partition lengths, or a partition length with a maximum value in the plurality of partition lengths may be used as the designated partition length;

Adjusting the enriched description vector sets corresponding to the other segmentation lengths except the specified segmentation length in the multiple segmentation lengths to form other enriched description vector sets with the same vector size as the specified enriched description vector set, wherein, for example, the vector sizes of the other enriched description vector sets can be stretched or compressed, such as screening or difference value, namely downsampling or upsampling, etc., so that other enriched description vector sets with the same vector size as the specified enriched description vector set can be formed;

Performing cascading combination operation on the appointed rich description vector set and the other rich description vector sets to form an aggregation key information description vector corresponding to the video data to be pushed, wherein the aggregation key information description vector can be { the appointed rich description vector set, the other rich description vector set 1, the other rich description vector set 2, the other rich description vector set 3, the other rich description vector set 4.

And carrying out pushing evaluation operation on the aggregated key information description vector so as to output pushing evaluation result data corresponding to the video data to be pushed, as described above.

It should be understood that, in some possible embodiments, the step of performing a cascade combination operation on the specified rich description vector set and the other rich description vector set to form an aggregate key information description vector corresponding to the video data to be pushed may further include the following specific implementation matters:

And performing cascading combination operation on row vector parameters or column vector parameters with identical component vector distribution coordinates in the specified enrichment description vector set and the other enrichment description vector sets to form a plurality of cascading combination vector parameter sets, and combining to form an aggregation key information description vector corresponding to the video data to be pushed based on the plurality of cascading combination vector parameter sets, for example, in the specified enrichment description vector set and the other enrichment description vector sets, cascading combination operation can be performed on row vector parameters of a first row or column vector parameters of a first column to form a first cascading combination vector parameter set, and cascading combination operation can be performed on row vector parameters of a second row or column vector parameters of a second column to form a second cascading combination vector parameter set.

With reference to fig. 3, the embodiment of the invention further provides an information pushing device based on big data, which can be applied to the information pushing system based on big data. Wherein, the big data based information pushing device may include some of the following software functional modules:

the data splitting module is used for respectively carrying out data splitting operation on the video data to be pushed based on each of a plurality of preset dividing lengths so as to form a video data fragment set corresponding to each of the dividing lengths, wherein for each dividing length, the video data fragment set corresponding to the dividing length comprises a plurality of video data fragments belonging to the dividing length, and each video data fragment comprises at least one frame of video frame to be pushed;

The key information mining module is used for respectively carrying out key information mining operation on a plurality of video data fragments in the video data fragment set corresponding to each division length so as to output a key information description vector set corresponding to each division length, wherein for each division length, the key information description vector set corresponding to the division length comprises key information description vectors mined by the video data fragments belonging to the division length;

The vector set splitting module is used for splitting the key information description vector set corresponding to the splitting length to form a plurality of key information description vector sub-sets corresponding to the splitting length, each key information description vector sub-set comprises at least one original key information description vector, and each original key information description vector corresponds to one video data segment;

The information enrichment module is used for carrying out information enrichment operation on original key information description vectors in the plurality of key information description vector sub-sets to form an enriched description vector sub-set corresponding to the plurality of key information description vector sub-sets, wherein the enriched description vector sub-set comprises at least one enriched key information description vector, and each enriched key information description vector corresponds to one original key information description vector in the enriched description vector sub-set;

The sub-set combination module is used for carrying out combination operation on the formed multiple enriched description vector sub-sets based on the precedence relation among the multiple key information description vector sub-sets so as to output the enriched description vector set corresponding to the segmentation length, wherein the video data fragment set, the key information description vector sub-set, the enriched description vector sub-set and the enriched description vector set all belong to an ordered set of description vectors;

The pushing evaluation module is used for carrying out aggregation operation on the enriched description vector set corresponding to each of the plurality of division lengths to form an aggregation key information description vector corresponding to the video data to be pushed, and carrying out pushing evaluation operation on the aggregation key information description vector to output pushing evaluation result data corresponding to the video data to be pushed, wherein the pushing evaluation result data is used for reflecting the pushing direction of carrying out data pushing operation on the video data to be pushed.

In summary, the method and the system for pushing information based on big data provided by the invention can split the video data to be pushed into a video data fragment set; performing key information mining operation on the video data fragment set, and outputting a key information description vector set; splitting the key information description vector set to form a key information description vector sub-set; carrying out information enrichment operation on the key information description vector sub-set to form an enriched description vector sub-set; carrying out combination operation on the enriched description vector sub-sets and outputting the enriched description vector sets; and carrying out aggregation operation on the enriched description vector set to form an aggregated key information description vector, carrying out push evaluation operation on the aggregated key information description vector, and outputting push evaluation result data. Based on the foregoing, since the aggregate key information description vector, which is the basis of the push evaluation operation, is the result of the information enrichment operation based on the video data segment set formed by performing the data splitting operation on the push video data with a plurality of division lengths, the aggregate key information description vector has stronger information characterization capability and higher reliability, so that the reliability of the determined video data push direction can be improved to a certain extent, the problem of low reliability of the determined video data push direction in the prior art is improved, the reliability of the video data push is improved to a certain extent, and the problem of poor reliability of the video data push in the prior art is improved.

The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. The information pushing method based on big data is characterized by comprising the following steps:

2. The big data based information pushing method of claim 1, wherein the step of performing an information enrichment operation on the original key information description vectors in the plurality of key information description vector sub-sets to form enriched description vector sub-sets corresponding to the plurality of key information description vector sub-sets comprises:

3. The big data based information pushing method of claim 1, wherein the step of performing an information enrichment operation on the original key information description vectors in the plurality of key information description vector sub-sets to form enriched description vector sub-sets corresponding to the plurality of key information description vector sub-sets comprises:

4. The method for pushing information based on big data as claimed in claim 3, wherein the step of performing a data integration operation on each of the key information description vector subsets and the distribution coordinates of the corresponding subset to output a rich description vector subset corresponding to each of the key information description vector subsets comprises:

5. The big data based information push method of claim 1, wherein each of the key information description vector subsets comprises a plurality of original key information description vectors;

6. The big data-based information pushing method of claim 5, wherein after the step of vector-combining the plurality of integrated key information description vectors on each of the same distribution coordinates in the formed plurality of integrated description vector sub-sets to form a corresponding one of the combined description vector sub-sets to output the corresponding plurality of combined description vector sub-sets, the big data-based information pushing method further comprises:

7. The big data based information pushing method of claim 1, wherein each of the subset of key information description vectors comprises a plurality of original key information description vectors, and any two of the subset of key information description vectors having an adjacency relationship comprise at least one identical original key information description vector;

8. The big data based information pushing method of claim 7, wherein the step of aggregating the rich key information description vectors corresponding to the same original key information description vector in each two rich description vector sub-sets having an adjacency relationship based on the precedence relationship between the plurality of key information description vector sub-sets comprises:

9. The big data based information pushing method according to any of the claims 1-8, wherein the set of enriched description vectors and the aggregated key information description vector both comprise row vector parameters and column vector parameters;

The step of performing an aggregation operation on the enriched description vector set corresponding to each of the plurality of partition lengths to form an aggregate key information description vector corresponding to the video data to be pushed, and performing a push evaluation operation on the aggregate key information description vector to output push evaluation result data corresponding to the video data to be pushed, includes:

10. A big data based information push system, comprising a processor and a memory, the memory being adapted to store a computer program, the processor being adapted to execute the computer program to implement the big data based information push method of any of claims 1-9.