CN120410629B - Advertisement structural feature reverse optimization generation system based on artificial intelligence - Google Patents

Advertisement structural feature reverse optimization generation system based on artificial intelligence

Info

Publication number
CN120410629B
CN120410629B CN202510928764.5A CN202510928764A CN120410629B CN 120410629 B CN120410629 B CN 120410629B CN 202510928764 A CN202510928764 A CN 202510928764A CN 120410629 B CN120410629 B CN 120410629B
Authority
CN
China
Prior art keywords
segment
advertisement
abnormal
information
advertising
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202510928764.5A
Other languages
Chinese (zh)
Other versions
CN120410629A (en
Inventor
赵文涛
肖鹏
曹豪杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Daoyoudao Technology Group Co ltd
Original Assignee
Daoyoudao Technology Group Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Daoyoudao Technology Group Co ltd filed Critical Daoyoudao Technology Group Co ltd
Priority to CN202510928764.5A priority Critical patent/CN120410629B/en
Publication of CN120410629A publication Critical patent/CN120410629A/en
Application granted granted Critical
Publication of CN120410629B publication Critical patent/CN120410629B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0277—Online advertisement
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0242—Determining effectiveness of advertisements
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0276—Advertisement creation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Multimedia (AREA)
  • Medical Informatics (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses an artificial intelligence-based advertisement structure feature reverse optimization generation system, which relates to the technical field of advertisement optimization, and comprises a historical advertisement disassembly module, an abnormal fragment analysis module, a fragment feature analysis module and a real-time advertisement construction module, wherein the historical advertisement disassembly module is used for generating corresponding advertisement pushing records, carrying out fragment disassembly and category division on advertisement video information, the abnormal fragment analysis module is used for identifying abnormal conditions of any advertisement fragment and carrying out abnormal feature extraction, the fragment feature analysis module is used for analyzing relevance between abnormal features and various advertisement fragments to obtain effective adaptation conditions between the abnormal features and various advertisement fragments, and the real-time advertisement construction module is used for carrying out abnormal feature extraction on advertisement information of target pushing advertisements to obtain expected advertisement fragment positions of corresponding information blocks and generating expected video structures.

Description

Advertisement structural feature reverse optimization generation system based on artificial intelligence
Technical Field
The invention relates to the technical field of advertisement optimization, in particular to an artificial intelligence-based advertisement structural feature reverse optimization generation system.
Background
Advertisement video optimization is a core carrier for optimizing advertisement delivery and improving user experience by utilizing artificial intelligence technology, and video advertisement has become a brand popularization in the current digital marketing environment. However, conventional advertising video production is highly dependent on manual experience, and still has a number of problems;
The differentiation rules of different flow videos are not quantified by a system, high flow videos often have specific structural characteristics, low flow videos have structural defects, for example, AI video generation software in the prior art focuses on general clips or templates in a multi-dimensional contrast analysis capability of the structural characteristics, reverse analysis structural models cannot be expressed based on the flow of historical advertisements, and therefore the utilization rate of a plurality of high-quality materials is low, and the expected feedback is difficult to obtain to cause the waste of resources.
Disclosure of Invention
The invention aims to provide an advertisement video pushing system and method based on artificial intelligence, which are used for solving the problems in the prior art.
In order to achieve the aim, the invention provides the technical scheme that the advertisement structure characteristic reverse optimization generation system based on artificial intelligence comprises a historical advertisement disassembly module, an abnormal fragment analysis module, a fragment characteristic analysis module and a real-time advertisement construction module;
The historical advertisement disassembly module is used for collecting data of advertisement videos pushed each time to generate corresponding advertisement pushing records, carrying out advertisement fragment disassembly on advertisement video information of any advertisement pushing record, and carrying out category classification on different advertisement fragments;
The abnormal segment analysis module is used for expanding multidimensional evaluation of pushing effect of any advertisement segment in the corresponding advertisement video, and identifying abnormal conditions of each advertisement segment;
the segment characteristic analysis module is used for analyzing the relevance between the abnormal characteristic and each advertisement segment based on the advertisement segment position condition of any abnormal characteristic in each advertisement segment;
the real-time advertisement construction module is used for carrying out information division on advertisement information of the target push advertisement and carrying out abnormal feature extraction, obtaining expected advertisement fragment positions of corresponding information blocks based on effective adaptation conditions between any abnormal feature and various advertisement fragments, and generating an expected video structure of the target push advertisement.
Further, the history advertisement disassembling module comprises a push record collecting unit and an advertisement fragment dividing unit;
The pushing record acquisition unit is used for extracting information of the content in the advertisement video pushed each time on the advertisement pushing platform to obtain a plurality of information elements of the advertisement video; presetting a plurality of information dimensions for the advertisement video, pre-matching an information set for each information dimension, arbitrarily selecting an information element to be compared with each information in each information set in similarity, and matching the selected information element with the compared information set to obtain the information dimension corresponding to the selected information element if the similarity exceeds a preset similarity threshold;
The information dimension of the advertisement video comprises product information, visual information, audio information, audience information and the like, and the information elements of the advertisement video comprise names, models, specifications and the like in the product information, colors, compositions, actions and the like in the visual information, sound effects, bystandings and the like in the audio information;
The advertisement segment dividing unit is used for acquiring the playing time interval of the advertisement video in any advertisement pushing record, extracting all information elements of each time point in the playing time interval to generate information labels of each time point, comparing the difference of the information labels of different time points, dividing the playing time interval into a plurality of advertisement segments of the advertisement video, and dividing all the advertisement segments into a plurality of categories based on the similarity of different advertisement segments on the information labels.
Further, the advertisement segment dividing unit includes:
Randomly selecting an advertisement push record and obtaining a playing time interval of a corresponding advertisement video, setting an initial time point when the advertisement video in the advertisement push record is selected to begin playing every time, and dividing the advertisement video every other unit time point to obtain a plurality of time points of the advertisement video;
randomly selecting a time point, extracting video information of the advertisement video at the selected time point, and obtaining a plurality of information elements of the selected time point; extracting information dimensions corresponding to each information element, summarizing the information dimensions corresponding to each information element, and generating an information label of a selected time point;
Randomly selecting two time points of the advertisement video, respectively extracting information labels of the two time points, comparing information elements of each information dimension in the information labels when the information labels of the two time points are the same, setting the two time points as similar time points if the information elements of the two time points in all the information dimensions are the same, setting the information elements of one information dimension as different time points if the information elements of one information dimension are different, merging the time points to obtain a target time interval if a plurality of adjacent and continuous time points are similar time points, setting the time points as an advertisement segment of the advertisement video, and generating a plurality of advertisement segments of the selected advertisement video;
Randomly selecting two advertisement fragments from the selected advertisement video, respectively extracting information labels in the two advertisement fragments, setting the two advertisement fragments as two types of advertisement fragments if different information dimensions exist in the two information labels, setting the two advertisement fragments as similar advertisement fragments if the information dimensions in the two information labels are the same, setting the information labels corresponding to the advertisement fragments as similar advertisement fragments, and generating a plurality of advertisement fragment types;
Judging whether two fragments are of the same type, only considering whether the information dimensions are the same, and allowing partial differences to exist on the information elements, such as different allowed sound effects in the audio information, but ensuring the sound effects of the same type of fragments, judging the same time interval, and judging that the information elements are the same, such as that the sound effects are changed, indicating that the advertisement information is changed.
Further, the abnormal fragment analysis module comprises an abnormal fragment identification unit and an abnormal feature extraction unit;
the advertisement segment identification unit is used for presetting a plurality of effect evaluation indexes for any advertisement segment to obtain an effect evaluation value under each effect evaluation index, obtaining a comprehensive evaluation value of any advertisement segment based on the pushing effect evaluation condition of each effect evaluation index, comparing the comprehensive evaluation value with a preset effect evaluation threshold, and judging whether the advertisement segment is an abnormal segment or not;
The abnormal feature extraction unit is used for arbitrarily selecting an abnormal segment, confirming the segment position of the selected abnormal segment, comparing the difference between the selected abnormal segment and the advertisement segment at the same segment position, extracting the abnormal feature of the selected abnormal segment, simultaneously comparing the difference between the selected abnormal segment and the advertisement segment of the same kind, and extracting the abnormal feature of the selected abnormal segment again.
Further, the abnormal section identifying unit includes:
When the advertisement pushing platform pushes the advertisement video, collecting pushing effects of the advertisement video through a plurality of preset monitoring modes, and setting corresponding effect evaluation indexes for each monitoring mode, randomly selecting an advertisement segment to obtain a time interval of selecting the advertisement segment, randomly selecting an ith effect evaluation index, extracting index values (a start)i and index data (a end)i) of an end time point of an initial time point from the time interval of selecting the advertisement segment, calculating to obtain an index change value delta a i of the ith effect evaluation index, and presetting an expected change value of the ith effect evaluation index as (delta a ex)i, and calculating to obtain an offset value of the ith effect evaluation index as p i=[Δai-(Δaex)i]/(Δaex)i;
And obtaining the deviation value of each effect evaluation index, summing to obtain a total deviation value P total of the selected advertisement fragments, presetting the effect evaluation threshold as P th, and setting the selected advertisement fragments as abnormal fragments if P total＞Pth.
Further, the abnormal feature extraction unit includes:
Randomly selecting an abnormal segment, extracting advertisement videos where the selected abnormal segment is located, obtaining the bit sequence of each advertisement segment according to the segment occurrence sequence, obtaining the segment position of the selected abnormal segment, setting the bit sequence of the selected abnormal segment as j, extracting two advertisement segments adjacent to the selected abnormal segment, and generating an advertisement segment group (A j-1,Aj,Aj+1) from three advertisement segments according to the bit sequence, wherein A j-1 is the advertisement segment with the bit sequence of j-1, A j is the advertisement segment with the bit sequence of j, and A j+1 is the advertisement segment with the bit sequence of j+1;
Selecting a reference record from the rest advertisement push records, generating a reference fragment group (A ' j-1,A' j,A' j+1) for three advertisement fragments with the same position in the selected reference record as the advertisement fragment group (A j-1,Aj,Aj+1), setting the advertisement fragment types of the advertisement fragments A j-1 and A j+1 and the relative position of the advertisement fragment A j+1 as the abnormal characteristics of the selected abnormal fragments if the advertisement fragment A j and the advertisement fragment A ' j are of the same type, comparing the information labels of the advertisement fragment A j and the advertisement fragment A ' j if the advertisement fragment A j and the advertisement fragment A ' j are not of the same type, and setting the information dimension different from the advertisement fragment A ' j in the advertisement fragment A j as the abnormal characteristics of the selected abnormal fragments;
Extracting an advertisement fragment set with the same type as the advertisement fragment of the selected abnormal fragment, arbitrarily selecting a reference fragment from the advertisement fragment set, comparing the information label of the selected abnormal fragment with the information label of the reference fragment, arbitrarily selecting one information dimension from the two information labels, comparing each information element in the selected information dimension, and setting different information elements as the abnormal characteristics of the selected abnormal fragment if the information elements different from the selected information dimension exist in the selected abnormal fragment;
The method comprises the steps of extracting abnormal characteristics of a selected abnormal segment and advertisement segments with the same segment position or the same advertisement segment type to generate an abnormal characteristic set of the selected abnormal segment, wherein one part of the abnormal characteristics in the advertisement segments are different characteristics existing in the same sequence, such as product display after brand information display, and have larger differences if background music display is performed after brand information display, and the other part of the abnormal characteristics are information element difference conditions under the same type of segment, such as obvious differences in effect caused by selection differences of background music.
Further, the segment characteristic analysis module comprises a segment characteristic association unit and a segment adaptation evaluation unit;
The segment characteristic association unit is used for analyzing the occurrence frequency of any abnormal characteristic in all advertisement segments and analyzing the segment position of any abnormal characteristic in each advertisement segment to obtain the association degree between each abnormal characteristic and different advertisement segments;
The segment adaptation evaluation unit is used for analyzing effect evaluation differences of any abnormal features in advertisement segments of the same category, and combining the association degrees between the abnormal features and different advertisement segments to obtain effective adaptation values between any abnormal features and the advertisement segments.
Further, the segment characteristic association unit includes:
Randomly selecting one abnormal feature, extracting all advertisement fragments containing the selected abnormal feature, setting the advertisement fragments as target advertisement fragments, obtaining the number of fragments of the target advertisement fragments as M, setting the total number of advertisement fragments in all advertisement pushing records as M total, and calculating to obtain the fragment number ratio of the selected abnormal feature as eta 1=M/M total;
The advertisement fragment types of each target advertisement fragment are obtained, the number of target advertisement fragments of the kth advertisement fragment is set to be N k, and the following formula is adopted: ;
And calculating to obtain a correlation value G k between the selected abnormal feature and the kth advertisement fragment, wherein the correlation value reflects the occurrence times of the abnormal feature in various advertisement fragments, and the larger the numerical value is, the larger the probability that the abnormal feature is placed in the advertisement fragment is.
Further, the segment adaptation evaluation unit includes:
Selecting another advertisement fragment from the same advertisement fragment type of the selected advertisement fragment to obtain the total deviation value of the another advertisement fragment as P ' total, obtaining the deviation difference value between the two advertisement fragments as delta P= |P total-P' total |, obtaining the difference feature quantity as r between the two advertisement fragments, and calculating to obtain the difference ratio tau=delta P/r of the selected abnormal feature on the two advertisement fragments;
Extracting the difference proportion of the selected abnormal feature on any two kth advertisement fragments, calculating the average value to obtain the average difference proportion tau k of the selected abnormal feature and the kth advertisement fragment, setting the association value G k between the selected abnormal feature and the kth advertisement fragment, and according to the formula: ;
And calculating to obtain an effective adaptation value S k between the selected abnormal characteristic and the kth advertisement segment.
Further, the real-time advertisement construction module comprises a target advertisement analysis unit and an expected structure generation unit;
The target advertisement analysis unit is used for acquiring information of a target push advertisement, extracting characteristics of the acquired information to obtain a plurality of information characteristics, and obtaining an information characteristic set of the target push advertisement;
The expected structure generating unit is used for randomly selecting one information feature from the information features, randomly selecting a kth advertisement fragment to obtain an effective adaptation value S k of the selected information feature and the kth advertisement fragment, respectively counting pushing frequencies of the kth advertisement fragment on different orders, setting the pushing frequency of the kth advertisement fragment on an xth order to be f (k,x), calculating an effective feature value T (k,x)=Sk×f(k,x) of the selected information feature in the kth advertisement fragment and in an xth dimension in the advertisement video, obtaining an expected advertisement fragment type and an expected fragment position of the selected information feature by selecting the effective feature value with the largest value of the selected information feature, and summarizing the expected advertisement fragment type and the expected fragment position of each information feature to generate an expected video structure of the target pushed advertisement.
Compared with the prior art, the invention has the beneficial effects that:
1. according to the invention, through structural disassembly and label comparison of historical advertisement videos and multidimensional evaluation of abnormal fragments, specific structural defects of low-efficiency fragments can be automatically identified from massive advertisement videos, and a quantization basis is provided for advertisement video optimization;
2. According to the invention, through introducing the relevance analysis of the abnormal characteristics and the advertisement fragments, the adaptability of the abnormal characteristics in different fragment types is verified by combining the historical data, so that the optimization strategy is ensured to adapt to different advertisement types, and the multiplexing rate of high-quality materials is obviously improved;
3. According to the invention, through a reverse optimization mechanism, the structural rule of the historical high-flow advertisement is converted into the generation rule, so that the manufacturing cost is reduced, the conversion rate of a user is improved, and the optimization level of the advertisement video can be effectively improved.
Drawings
FIG. 1 is a schematic diagram of an artificial intelligence based advertisement structural feature reverse optimization generation system.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. 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 invention provides an artificial intelligence-based advertisement structure feature reverse optimization generation system, which comprises a historical advertisement disassembly module, an abnormal fragment analysis module, a fragment feature analysis module and a real-time advertisement construction module;
The historical advertisement disassembly module is used for collecting data of advertisement videos pushed each time to generate corresponding advertisement pushing records, carrying out advertisement fragment disassembly on advertisement video information of any advertisement pushing record, and carrying out category classification on different advertisement fragments;
the history advertisement disassembling module comprises a push record collecting unit and an advertisement fragment dividing unit;
The pushing record acquisition unit is used for extracting information of the content in the advertisement video pushed each time on the advertisement pushing platform to obtain a plurality of information elements of the advertisement video; presetting a plurality of information dimensions for the advertisement video, pre-matching an information set for each information dimension, arbitrarily selecting an information element to be compared with each information in each information set in similarity, and matching the selected information element with the compared information set to obtain the information dimension corresponding to the selected information element if the similarity exceeds a preset similarity threshold;
The advertisement segment dividing unit is used for acquiring the playing time interval of the advertisement video in any advertisement pushing record, extracting all information elements of each time point in the playing time interval to generate information labels of each time point, comparing the difference of the information labels of different time points, dividing the playing time interval into a plurality of advertisement segments of the advertisement video, and dividing all the advertisement segments into a plurality of categories based on the similarity of different advertisement segments on the information labels.
Wherein, advertising segment dividing unit includes:
Randomly selecting an advertisement push record and obtaining a playing time interval of a corresponding advertisement video, setting an initial time point when the advertisement video in the advertisement push record is selected to begin playing every time, and dividing the advertisement video every other unit time point to obtain a plurality of time points of the advertisement video;
randomly selecting a time point, extracting video information of the advertisement video at the selected time point, and obtaining a plurality of information elements of the selected time point; extracting information dimensions corresponding to each information element, summarizing the information dimensions corresponding to each information element, and generating an information label of a selected time point;
Randomly selecting two time points of the advertisement video, respectively extracting information labels of the two time points, comparing information elements of each information dimension in the information labels when the information labels of the two time points are the same, setting the two time points as similar time points if the information elements of the two time points in all the information dimensions are the same, setting the information elements of one information dimension as different time points if the information elements of one information dimension are different, merging the time points to obtain a target time interval if a plurality of adjacent and continuous time points are similar time points, setting the time points as an advertisement segment of the advertisement video, and generating a plurality of advertisement segments of the selected advertisement video;
And randomly selecting two advertisement fragments from the selected advertisement video, respectively extracting information labels in the two advertisement fragments, setting the two advertisement fragments as two types of advertisement fragments if different information dimensions exist in the two information labels, setting the two advertisement fragments as similar advertisement fragments if the information dimensions in the two information labels are the same, setting the information labels corresponding to the advertisement fragments as the type labels of the similar advertisement fragments, and generating a plurality of advertisement fragment types.
The abnormal segment analysis module is used for expanding multidimensional evaluation of pushing effect of any advertisement segment in the corresponding advertisement video, and identifying abnormal conditions of each advertisement segment;
The abnormal fragment analysis module comprises an abnormal fragment identification unit and an abnormal feature extraction unit;
the advertisement segment identification unit is used for presetting a plurality of effect evaluation indexes for any advertisement segment to obtain an effect evaluation value under each effect evaluation index, obtaining a comprehensive evaluation value of any advertisement segment based on the pushing effect evaluation condition of each effect evaluation index, comparing the comprehensive evaluation value with a preset effect evaluation threshold, and judging whether the advertisement segment is an abnormal segment or not;
The abnormal feature extraction unit is used for arbitrarily selecting an abnormal segment, confirming the segment position of the selected abnormal segment, comparing the difference between the selected abnormal segment and the advertisement segment at the same segment position, extracting the abnormal feature of the selected abnormal segment, simultaneously comparing the difference between the selected abnormal segment and the advertisement segment of the same kind, and extracting the abnormal feature of the selected abnormal segment again.
Wherein the abnormal section identifying unit includes:
When the advertisement pushing platform pushes the advertisement video, collecting pushing effects of the advertisement video through a plurality of preset monitoring modes, and setting corresponding effect evaluation indexes for each monitoring mode, randomly selecting an advertisement segment to obtain a time interval of selecting the advertisement segment, randomly selecting an ith effect evaluation index, extracting index values (a start)i and index data (a end)i) of an end time point of an initial time point from the time interval of selecting the advertisement segment, calculating to obtain an index change value delta a i of the ith effect evaluation index, and presetting an expected change value of the ith effect evaluation index as (delta a ex)i, and calculating to obtain an offset value of the ith effect evaluation index as p i=[Δai-(Δaex)i]/(Δaex)i;
And obtaining the deviation value of each effect evaluation index, summing to obtain a total deviation value P total of the selected advertisement fragments, presetting the effect evaluation threshold as P th, and setting the selected advertisement fragments as abnormal fragments if P total＞Pth.
Wherein the abnormal feature extraction unit includes:
Randomly selecting an abnormal segment, extracting advertisement videos where the selected abnormal segment is located, obtaining the bit sequence of each advertisement segment according to the segment occurrence sequence, obtaining the segment position of the selected abnormal segment, setting the bit sequence of the selected abnormal segment as j, extracting two advertisement segments adjacent to the selected abnormal segment, and generating an advertisement segment group (A j-1,Aj,Aj+1) from three advertisement segments according to the bit sequence, wherein A j-1 is the advertisement segment with the bit sequence of j-1, A j is the advertisement segment with the bit sequence of j, and A j+1 is the advertisement segment with the bit sequence of j+1;
Selecting a reference record from the rest advertisement push records, generating a reference fragment group (A ' j-1,A' j,A' j+1) for three advertisement fragments with the same position in the selected reference record as the advertisement fragment group (A j-1,Aj,Aj+1), setting the advertisement fragment types of the advertisement fragments A j-1 and A j+1 and the relative position of the advertisement fragment A j+1 as the abnormal characteristics of the selected abnormal fragments if the advertisement fragment A j and the advertisement fragment A ' j are of the same type, comparing the information labels of the advertisement fragment A j and the advertisement fragment A ' j if the advertisement fragment A j and the advertisement fragment A ' j are not of the same type, and setting the information dimension different from the advertisement fragment A ' j in the advertisement fragment A j as the abnormal characteristics of the selected abnormal fragments;
Extracting an advertisement fragment set with the same type as the advertisement fragment of the selected abnormal fragment, arbitrarily selecting a reference fragment from the advertisement fragment set, comparing the information label of the selected abnormal fragment with the information label of the reference fragment, arbitrarily selecting one information dimension from the two information labels, comparing each information element in the selected information dimension, and setting different information elements as the abnormal characteristics of the selected abnormal fragment if the information elements different from the selected information dimension exist in the selected abnormal fragment;
And carrying out abnormal feature extraction on the selected abnormal fragments and the advertisement fragments with the same fragment positions or the same advertisement fragment types, and generating an abnormal feature set of the selected abnormal fragments.
The segment characteristic analysis module is used for analyzing the relevance between the abnormal characteristic and each advertisement segment based on the advertisement segment position condition of any abnormal characteristic in each advertisement segment;
the fragment characteristic analysis module comprises a fragment characteristic association unit and a fragment adaptation evaluation unit;
The segment characteristic association unit is used for analyzing the occurrence frequency of any abnormal characteristic in all advertisement segments and analyzing the segment position of any abnormal characteristic in each advertisement segment to obtain the association degree between each abnormal characteristic and different advertisement segments;
The segment adaptation evaluation unit is used for analyzing effect evaluation differences of any abnormal features in advertisement segments of the same category, and combining the association degrees between the abnormal features and different advertisement segments to obtain effective adaptation values between any abnormal features and the advertisement segments.
Wherein, fragment characteristic association unit includes:
Randomly selecting one abnormal feature, extracting all advertisement fragments containing the selected abnormal feature, setting the advertisement fragments as target advertisement fragments, obtaining the number of fragments of the target advertisement fragments as M, setting the total number of advertisement fragments in all advertisement pushing records as M total, and calculating to obtain the fragment number ratio of the selected abnormal feature as eta 1=M/M total;
The advertisement fragment types of each target advertisement fragment are obtained, the number of target advertisement fragments of the kth advertisement fragment is set to be N k, and the following formula is adopted: ;
Calculating to obtain a correlation value G k between the selected abnormal feature and the kth advertisement segment;
selecting an abnormal feature as bystanders, obtaining 50 advertisement fragments containing the bystanders, setting the total number of fragments in all advertisement pushing records as 100, obtaining the proportion of the fragments of the bystanders as 50%, counting the number of the advertisement fragments as 20, and calculating to obtain the association value G=20/50×50% =20% between the bystanders and the openings;
wherein the segment adaptation evaluation unit comprises:
Selecting another advertisement fragment from the same advertisement fragment type of the selected advertisement fragment to obtain the total deviation value of the another advertisement fragment as P ' total, obtaining the deviation difference value between the two advertisement fragments as delta P= |P total-P' total |, obtaining the difference feature quantity as r between the two advertisement fragments, and calculating to obtain the difference ratio tau=delta P/r of the selected abnormal feature on the two advertisement fragments;
Extracting the difference proportion of the selected abnormal feature on any two kth advertisement fragments, calculating the average value to obtain the average difference proportion tau k of the selected abnormal feature and the kth advertisement fragment, setting the association value G k between the selected abnormal feature and the kth advertisement fragment, and according to the formula: ;
Calculating to obtain an effective adaptation value S k between the selected abnormal characteristic and the kth advertisement segment;
Embodiment 2 setting the deviation difference value of the bypass in two advertisement fragments as 20%, simultaneously obtaining the difference characteristic quantity between the two advertisement fragments as 5, obtaining the difference ratio of the bypass on the two advertisement fragments as 4%, averaging the difference ratio of any advertisement fragments to obtain the average difference ratio as 5%, setting the association value of the bypass and the opening as 20%, and calculating to obtain the effective adaptation value S=20% × (1-5%) =19%.
The real-time advertisement construction module is used for carrying out information division on advertisement information of the target push advertisement and carrying out abnormal feature extraction, obtaining expected advertisement fragment positions of corresponding information blocks based on effective adaptation conditions between any abnormal feature and various advertisement fragments, and generating an expected video structure of the target push advertisement;
the real-time advertisement construction module comprises a target advertisement analysis unit and an expected structure generation unit;
The target advertisement analysis unit is used for acquiring information of a target push advertisement, extracting characteristics of the acquired information to obtain a plurality of information characteristics, and obtaining an information characteristic set of the target push advertisement;
The expected structure generating unit is used for randomly selecting one information feature from the information features, randomly selecting a kth advertisement fragment to obtain an effective adaptation value S k of the selected information feature and the kth advertisement fragment, respectively counting pushing frequencies of the kth advertisement fragment on different orders, setting the pushing frequency of the kth advertisement fragment on an xth order to be f (k,x), calculating an effective feature value T (k,x)=Sk×f(k,x) of the selected information feature in the kth advertisement fragment and in an xth dimension in the advertisement video, obtaining an expected advertisement fragment type and an expected fragment position of the selected information feature by selecting the effective feature value with the largest value of the selected information feature, and summarizing the expected advertisement fragment type and the expected fragment position of each information feature to generate an expected video structure of the target pushed advertisement.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Claims (9)

1.一种基于人工智能的广告结构特征反向优化生成系统，其特征在于：所述生成系统包括历史广告拆解模块、异常片段分析模块、片段特征分析模块和实时广告构建模块；1. An AI-based advertising structure feature reverse optimization generation system, characterized in that: the generation system includes a historical advertising decomposition module, an abnormal fragment analysis module, a fragment feature analysis module, and a real-time advertising construction module; 所述历史广告拆解模块，用于对每一次推送的广告视频进行数据采集，生成相应的广告推送记录；对任意广告推送记录的广告视频信息进行广告片段拆解，并对不同广告片段进行种类划分；The historical ad breakdown module is used to collect data from each pushed ad video and generate corresponding ad push records; to break down ad video information in any ad push record into ad segments and to classify different ad segments into categories. 所述异常片段分析模块，用于对任意广告片段在相应广告视频中的推送效果展开多维度评估，对各个广告片段的异常情况进行识别；基于广告片段位置差异和效果评估差异，对任意异常广告片段进行异常特征提取；The abnormal segment analysis module is used to conduct multi-dimensional evaluation of the push effect of any advertising segment in the corresponding advertising video, identify the abnormality of each advertising segment, and extract abnormal features of any abnormal advertising segment based on the differences in the position and the differences in the effect evaluation of the advertising segments. 所述片段特征分析模块，用于基于任意异常特征在各个广告片段中的广告片段位置情况，对异常特征与各类广告片段之间的关联性进行分析；根据任意异常特征在各类广告片段中的效果评估情况，得到异常特征与各类广告片段之间的有效适应情况；The segment feature analysis module is used to analyze the correlation between abnormal features and various types of advertising segments based on the position of any abnormal feature in each advertising segment; and to obtain the effective adaptation between abnormal features and various types of advertising segments based on the effect evaluation of any abnormal feature in various types of advertising segments. 所述实时广告构建模块，用于对目标推送广告的广告信息进行信息划分并进行异常特征提取；基于任意异常特征与各类广告片段之间的有效适应情况，得到相应信息块的期望广告片段位置，并生成目标推送广告的期望视频结构；The real-time advertising construction module is used to divide the advertising information of the target push advertisement and extract abnormal features; based on the effective adaptation between any abnormal features and various types of advertising segments, the expected advertising segment position of the corresponding information block is obtained, and the expected video structure of the target push advertisement is generated. 所述异常片段分析模块包括异常片段识别单元和异常特征提取单元；The abnormal fragment analysis module includes an abnormal fragment identification unit and an abnormal feature extraction unit; 所述异常特征提取单元，包括：The abnormal feature extraction unit includes: 任意选取一个异常片段，对选取异常片段所在广告视频进行提取，按照片段发生顺序得到各个广告片段的位序，得到选取异常片段的片段位置，设定选取异常片段的位序为j，将与选取异常片段相邻的两个广告片段进行提取，将三个广告片段按照位序生成广告片段组(Aj-1,Aj,Aj+1)，其中，Aj-1为位序为j-1的广告片段，Aj为位序为j的广告片段，Aj+1为位序为j+1的广告片段；Arbitrarily select an abnormal segment, extract the advertisement video containing the selected abnormal segment, obtain the position of each advertisement segment according to the order in which the segments occur, obtain the position of the selected abnormal segment, set the position of the selected abnormal segment as j, extract the two advertisement segments adjacent to the selected abnormal segment, and generate an advertisement segment group (Aj -1 , Aj , Aj +1 ) according to the position of the three advertisement segments, where Aj-1 is the advertisement segment with position j-1, Aj is the advertisement segment with position j, and Aj+1 is the advertisement segment with position j+1. 从其余广告推送记录中任意选取一条参照记录，对选取参照记录中位序与广告片段组(Aj-1,Aj,Aj+1)相同的三个广告片段，生成一个参照片段组(A’ j-1,A’ j,A’ j+1)；若广告片段Aj与广告片段A’ j为相同类型，则将广告片段Aj-1和广告片段Aj+1的广告片段类型和与选取异常片段的相对位序设定为选取异常片段的异常特征；若广告片段Aj与广告片段A’ j不为相同类型，则对广告片段Aj与广告片段A’ j的信息标签进行比对，将广告片段Aj中，与广告片段A’ j不同的信息维度设定为选取异常片段的异常特征；From the remaining ad push records, arbitrarily select a reference record. For the three ad segments in the selected reference record whose position is the same as the ad segment group ( Aj-1 , Aj , Aj +1 ), generate a reference segment group ( A'j -1 , A'j , A'j +1 ) . If ad segment Aj and ad segment A'j are of the same type, then set the ad segment type of ad segment Aj-1 and ad segment Aj +1 and their relative position with the selected abnormal segment as the abnormal feature of the selected abnormal segment. If ad segment Aj and ad segment A'j are not of the same type, then compare the information tags of ad segment Aj and ad segment A'j , and set the information dimensions in ad segment Aj that are different from ad segment A'j as the abnormal feature of the selected abnormal segment. 所述片段特征分析模块包括片段特征关联单元和片段适应评估单元；The fragment feature analysis module includes a fragment feature association unit and a fragment adaptation evaluation unit; 所述片段特征关联单元，包括：The fragment feature association unit includes: 任意选取一个异常特征，提取包含有选取异常特征的所有广告片段并设定为目标广告片段，得到所述目标广告片段的片段数量为M，设定所有广告推送记录中的广告片段总数量为Mtotal，计算得到选取异常特征的片段数量占比为η1=M/Mtotal；Arbitrarily select an abnormal feature, extract all ad segments containing the selected abnormal feature and set them as target ad segments, and obtain the number of target ad segments as M. Set the total number of ad segments in all ad push records as M total , and calculate the proportion of the number of segments with selected abnormal features as η1 = M / M total . 获取各个目标广告片段的广告片段种类，设定第k种广告片段的目标广告片段数量为Nk，根据公式：Obtain the ad segment types for each target ad segment, and set the number of target ad segments of the k-th type to N <sub>k</sub> , according to the formula: ； ; 计算得到选取异常特征与第k种广告片段之间的关联值Gk。The correlation value G <sub>k</sub> between the selected abnormal features and the k-th type of advertisement segment is calculated. 2.根据权利要求1所述的一种基于人工智能的广告结构特征反向优化生成系统，其特征在于：所述历史广告拆解模块包括推送记录采集单元和广告片段划分单元；2. The AI-based advertising structure feature reverse optimization generation system according to claim 1, characterized in that: the historical advertising decomposition module includes a push record collection unit and an advertising segment division unit; 所述推送记录采集单元，用于对广告推送平台上每一次推送的广告视频中的内容进行信息提取，得到广告视频的若干个信息元；对广告视频预设若干个信息维度，对每一个信息维度预先匹配一个信息集合，任意选取一个信息元与各个信息集合中的各个信息进行相似度比对，若相似度超过预设的相似度阈值，则将选取信息元与比对的信息集合进行匹配，得到选取信息元对应的信息维度；将各个信息元与信息维度进行匹配，得到各个信息维度的信息元集合，生成相应的广告推送记录；The push record collection unit is used to extract information from the content of each push advertisement video on the advertising push platform to obtain several information elements of the advertisement video; to preset several information dimensions for the advertisement video, to pre-match an information set for each information dimension, to arbitrarily select an information element and compare its similarity with each piece of information in each information set, and if the similarity exceeds a preset similarity threshold, to match the selected information element with the compared information set to obtain the information dimension corresponding to the selected information element; to match each information element with the information dimension to obtain the information element set of each information dimension, and to generate the corresponding advertisement push record; 所述广告片段划分单元，用于获取任意广告推送记录中广告视频的播放时间区间，并对播放时间区间中各个时间点的所有信息元进行提取，生成各个时间点的信息标签；对不同时间点的信息标签进行差异比对，将播放时间区间划分为广告视频的若干个广告片段，并基于不同广告片段在信息标签上的相似情况，将所有广告片段划分为若干个种类。The advertising segment segmentation unit is used to obtain the playback time interval of the advertising video in any advertising push record, extract all information elements at each time point in the playback time interval, and generate information tags for each time point; compare the differences of information tags at different time points, divide the playback time interval into several advertising segments of the advertising video, and classify all advertising segments into several categories based on the similarity of information tags of different advertising segments. 3.根据权利要求2所述的一种基于人工智能的广告结构特征反向优化生成系统，其特征在于：所述广告片段划分单元，包括：3. The advertising structure feature reverse optimization generation system based on artificial intelligence according to claim 2, characterized in that: the advertising segment division unit includes: 任意选取一条广告推送记录并得到对应广告视频的播放时间区间，每当选取广告推送记录中的广告视频开始播放时，则设置为初始时间点，并每隔一个单位时间点对广告视频进行划分，得到广告视频的若干个时间点；Arbitrarily select an ad push record and obtain the playback time interval of the corresponding ad video. Whenever an ad video in the selected ad push record starts playing, it is set as the initial time point. The ad video is divided into several time points every unit of time. 任意选取一个时间点，对广告视频在选取时间点下的视频信息进行提取，得到选取时间点的若干个信息元；提取每一个信息元对应的信息维度，将各个信息元对应的信息维度进行汇总，生成选取时间点的一个信息标签；Select any time point and extract the video information of the advertisement video at the selected time point to obtain several information elements of the selected time point; extract the information dimension corresponding to each information element, summarize the information dimensions corresponding to each information element, and generate an information tag for the selected time point; 随机选取广告视频的两个时间点，分别对两个时间点的信息标签进行提取，当两个时间点的信息标签相同，则对信息标签中各个信息维度的信息元进行比对，若两个时间点在所有信息维度的信息元均相同，则将两个时间点设定为同类时间点，若存在一个信息维度的信息元不相同，则设定为异类时间点；若存在彼此相邻且连续的若干个时间点互为同类时间点，则将若干个时间点合并得到一个目标时间区间，并设定为广告视频的一个广告片段，生成选取广告视频的若干个广告片段；Two time points are randomly selected from the advertisement video. Information tags are extracted from the two time points. If the information tags of the two time points are the same, the information elements of each information dimension in the information tags are compared. If the information elements of the two time points are the same in all information dimensions, the two time points are set as the same type of time points. If the information elements of one information dimension are different, they are set as different types of time points. If there are several adjacent and consecutive time points that are of the same type, the several time points are merged to obtain a target time interval, which is set as an advertisement segment of the advertisement video, and several advertisement segments of the selected advertisement video are generated. 从选取广告视频中任意选取两个广告片段，分别提取两个广告片段中的信息标签，若两个信息标签中存在不同的信息维度，则将两个广告片段设定为两个种类广告片段，若两个信息标签中各个信息维度均相同，则将两个广告片段设定为同类广告片段，并将广告片段对应的信息标签设定为同类广告片段的种类标签，生成若干种广告片段类型。Select any two ad segments from the selected ad videos, and extract the information tags from the two ad segments respectively. If the two information tags have different information dimensions, then the two ad segments are set as two different types of ad segments. If all information dimensions in the two information tags are the same, then the two ad segments are set as the same type of ad segments, and the information tags corresponding to the ad segments are set as the category tags of the same type of ad segments, generating several types of ad segments. 4.根据权利要求3所述的一种基于人工智能的广告结构特征反向优化生成系统，其特征在于：所述异常片段分析模块包括异常片段识别单元和异常特征提取单元；4. The advertising structure feature reverse optimization generation system based on artificial intelligence according to claim 3, characterized in that: the abnormal fragment analysis module includes an abnormal fragment identification unit and an abnormal feature extraction unit; 所述异常片段识别单元，用于对任意广告片段预设若干种效果评估指标，得到每一种效果评估指标下的效果评估值；基于各个效果评估指标的推送效果评估情况，得到任意广告片段的综合评估值，并与预设的效果评估阈值进行比对，判断广告片段是否为异常片段；The abnormal segment identification unit is used to preset several effect evaluation indicators for any advertising segment, obtain the effect evaluation value under each effect evaluation indicator; based on the push effect evaluation of each effect evaluation indicator, obtain the comprehensive evaluation value of any advertising segment, and compare it with the preset effect evaluation threshold to determine whether the advertising segment is an abnormal segment. 所述异常特征提取单元，用于任意选取一个异常片段，并对选取异常片段的片段位置进行确认；将选取异常片段与相同片段位置的广告片段进行差异比对，对选取异常片段进行异常特征提取，同时将选取异常片段与同类广告片段进行差异比对，对选取异常片段再次进行异常特征提取。The abnormal feature extraction unit is used to arbitrarily select an abnormal segment and confirm the segment position of the selected abnormal segment; compare the selected abnormal segment with advertising segments at the same segment position to extract abnormal features from the selected abnormal segment; and compare the selected abnormal segment with advertising segments of the same type to extract abnormal features from the selected abnormal segment again. 5.根据权利要求4所述的一种基于人工智能的广告结构特征反向优化生成系统，其特征在于：所述异常片段识别单元，包括：5. The AI-based advertising structure feature reverse optimization generation system according to claim 4, characterized in that: the abnormal fragment identification unit comprises: 当广告推送平台推送广告视频后，通过预设的若干种监测方式对广告视频的推送效果进行采集，对每一种监测方式设定相应的效果评估指标；任意选取一个广告片段，得到选取广告片段的时间区间，任意选取第i个效果评估指标，从选取广告片段的时间区间中提取初始时间点的指标数值(astart)i和末端时间点的指标数据(aend)i，计算得到第i个效果评估指标的指标变化值Δai，预设第i个效果评估指标的期望变化值为(Δaex)i，计算得到第i个效果评估指标的偏差值为pi=[Δai-(Δaex)i]/(Δaex)i；After the advertising platform pushes an advertising video, it collects the push effect of the advertising video through several preset monitoring methods, and sets corresponding effect evaluation indicators for each monitoring method; it arbitrarily selects an advertising segment, obtains the time interval of the selected advertising segment, arbitrarily selects the i-th effect evaluation indicator, extracts the indicator value (a start ) i at the initial time point and the indicator data (a end ) i at the end time point from the time interval of the selected advertising segment, calculates the indicator change value Δa i of the i-th effect evaluation indicator, presets the expected change value of the i-th effect evaluation indicator as (Δa ex ) i , and calculates the deviation value of the i-th effect evaluation indicator as p i = [Δa i - (Δa ex ) i ]/(Δa ex ) i ; 对每个效果评估指标的偏差值进行获取，进行求和得到选取广告片段的偏差总值Ptotal，预设效果评估阈值为Pth，若Ptotal＞Pth，则将选取广告片段设定为异常片段。The deviation values of each performance evaluation indicator are obtained and summed to obtain the total deviation value P_total of the selected advertising segments. The preset performance evaluation threshold is P_th . If P_total > P_th , the selected advertising segment is set as an abnormal segment. 6.根据权利要求5所述的一种基于人工智能的广告结构特征反向优化生成系统，其特征在于：所述异常特征提取单元，还包括：6. The advertising structure feature reverse optimization generation system based on artificial intelligence according to claim 5, characterized in that: the abnormal feature extraction unit further includes: 提取与选取异常片段的广告片段类型相同的广告片段集合，从所述广告片段集合中任意选取一个参照片段，将选取异常片段的信息标签与所述参照片段的信息标签进行比对，从两个信息标签中任意选取一个信息维度，将选取信息维度中的各个信息元进行比对，若选取异常片段中存在与选取信息维度不同的信息元，则将不同的信息元设定为选取异常片段的异常特征；Extract a set of ad segments of the same type as the selected abnormal segment. Randomly select a reference segment from the set of ad segments. Compare the information tags of the selected abnormal segment with the information tags of the reference segment. Randomly select an information dimension from the two information tags. Compare each information element in the selected information dimension. If there are information elements in the selected abnormal segment that are different from the selected information dimension, then set the different information elements as the abnormal features of the selected abnormal segment. 将选取异常片段与相同片段位置或者相同广告片段类型的广告片段进行异常特征提取，生成选取异常片段的异常特征集合。The abnormal segments are compared with ad segments in the same position or of the same ad segment type to extract abnormal features, generating an abnormal feature set of the selected abnormal segments. 7.根据权利要求6所述的一种基于人工智能的广告结构特征反向优化生成系统，其特征在于：所述片段特征分析模块包括片段特征关联单元和片段适应评估单元；7. The advertising structure feature reverse optimization generation system based on artificial intelligence according to claim 6, characterized in that: the segment feature analysis module includes a segment feature association unit and a segment adaptation evaluation unit; 所述片段特征关联单元，用于对任意异常特征在所有广告片段中的发生频率进行分析，并对任意异常特征在各个广告片段中的片段位置进行分析，得到各个异常特征与不同广告片段之间的关联程度；The segment feature association unit is used to analyze the frequency of occurrence of any abnormal feature in all advertising segments and to analyze the segment position of any abnormal feature in each advertising segment, so as to obtain the degree of association between each abnormal feature and different advertising segments. 所述片段适应评估单元，用于对任意异常特征在相同类别的广告片段中的效果评估差异进行分析，结合异常特征与不同广告片段之间的关联程度，得到任意异常特征与该类广告片段之间的有效适应值。The segment adaptation evaluation unit is used to analyze the difference in the effectiveness evaluation of any anomalous feature in advertising segments of the same category, and to obtain the effective adaptation value between any anomalous feature and advertising segments of that category by combining the degree of correlation between the anomalous feature and different advertising segments. 8.根据权利要求7所述的一种基于人工智能的广告结构特征反向优化生成系统，其特征在于：所述片段适应评估单元，包括：8. The advertising structure feature reverse optimization generation system based on artificial intelligence according to claim 7, characterized in that: the segment adaptation evaluation unit includes: 任意选取一个异常特征，从第k种广告片段中任意选取一个包含有选取异常特征的广告片段，得到选取异常特征所在选取广告片段的偏差总值为Ptotal；从选取广告片段的相同广告片段类型中选取另外一个广告片段，得到所述另外一个广告片段的偏差总值为P’ total，得到两个广告片段之间的偏差差值为ΔP=|Ptotal-P’ total|；获取两个广告片段之间的差异特征数量为r，计算得到选取异常特征在两个广告片段上的差异占比τ=ΔP/r；Arbitrarily select an anomalous feature, and arbitrarily select an ad segment containing the selected anomalous feature from the k-th ad segment. Obtain the total deviation value P_total of the selected ad segment containing the selected anomalous feature. Select another ad segment from the same ad segment type of the selected ad segment, and obtain the total deviation value P'_total of the other ad segment. Obtain the deviation difference between the two ad segments as ΔP = | P_total - P'_total |. Obtain the number of difference features between the two ad segments as r, and calculate the difference ratio τ = ΔP/r of the selected anomalous feature between the two ad segments. 对选取异常特征在任意两个第k种广告片段上的差异占比进行提取，进行平均值计算得到选取异常特征与第k种广告片段的平均差异占比τk；设定选取异常特征与第k种广告片段之间的关联值Gk，根据公式：Extract the percentage difference between the selected anomaly feature and any two ad segments of type k, and calculate the average percentage difference τk between the selected anomaly feature and the ad segment of type k. Define the correlation value Gk between the selected anomaly feature and the ad segment of type k, according to the formula: ； ; 计算得到选取异常特征与第k种广告片段之间的有效适应值Sk。The effective fitness value Sk between the selected abnormal features and the k-th advertising segment is calculated. 9.根据权利要求8所述的一种基于人工智能的广告结构特征反向优化生成系统，其特征在于：所述实时广告构建模块包括目标广告分析单元和期望结构生成单元；9. The AI-based advertising structure feature reverse optimization generation system according to claim 8, characterized in that: the real-time advertising construction module includes a target advertising analysis unit and a desired structure generation unit; 所述目标广告分析单元，用于对目标推送广告进行信息获取，并对获取的信息进行特征提取得到若干个信息特征，得到所述目标推送广告的信息特征集合；将所述信息特征集合中的任意一个信息特征与各种异常片段的异常特征集合进行比对，生成各种异常片段的实时异常特征集合；The target advertisement analysis unit is used to acquire information about the target push advertisement, extract features from the acquired information to obtain several information features, and obtain the information feature set of the target push advertisement; compare any one information feature in the information feature set with the abnormal feature set of various abnormal segments to generate a real-time abnormal feature set of various abnormal segments. 所述期望结构生成单元，用于从所述若干个信息特征中任意选取一个信息特征，并任意选取第k种广告片段，得到选取信息特征与第k种广告片段的有效适应值Sk；分别统计第k种广告片段在不同位序上的推送频率，设定第k种广告片段在第x个位序上的推送频率为f(k,x)，计算得到选取信息特征处于第k种广告片段时，并且在广告视频中处于第x个维度的有效特征值T(k,x)=Sk×f(k,x)；对选取信息特征选取数值最大的一个有效特征值，得到选取信息特征的期望广告片段种类和期望片段位置；对各个信息特征的期望广告片段种类和期望片段位置进行汇总，生成目标推送广告的期望视频结构。The desired structure generation unit is used to arbitrarily select one information feature from the plurality of information features, and arbitrarily select the k-th type of advertising segment to obtain the effective adaptation value Sk of the selected information feature and the k-th type of advertising segment; to count the push frequency of the k-th type of advertising segment in different positions, and set the push frequency of the k-th type of advertising segment in the x-th position as f (k,x) , and calculate the effective feature value T (k,x) = Sk ×f (k,x) when the selected information feature is in the k-th type of advertising segment and in the x-th dimension of the advertising video; to select the effective feature value with the largest value for the selected information feature, and obtain the desired advertising segment type and desired segment position of the selected information feature; to summarize the desired advertising segment type and desired segment position of each information feature, and generate the desired video structure of the target push advertisement.
CN202510928764.5A 2025-07-07 2025-07-07 Advertisement structural feature reverse optimization generation system based on artificial intelligence Active CN120410629B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202510928764.5A CN120410629B (en) 2025-07-07 2025-07-07 Advertisement structural feature reverse optimization generation system based on artificial intelligence

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202510928764.5A CN120410629B (en) 2025-07-07 2025-07-07 Advertisement structural feature reverse optimization generation system based on artificial intelligence

Publications (2)

Publication Number Publication Date
CN120410629A CN120410629A (en) 2025-08-01
CN120410629B true CN120410629B (en) 2025-12-09

Family

ID=96517870

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202510928764.5A Active CN120410629B (en) 2025-07-07 2025-07-07 Advertisement structural feature reverse optimization generation system based on artificial intelligence

Country Status (1)

Country Link
CN (1) CN120410629B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119046481A (en) * 2024-11-01 2024-11-29 融安云网(北京)技术有限公司 Multimedia video stream management system and method based on artificial intelligence
CN120166238A (en) * 2025-03-20 2025-06-17 南京数海星辰视觉科技有限公司 Video content analysis and post-production optimization method and system based on cloud computing
CN120198178A (en) * 2025-05-26 2025-06-24 长春慧程科技有限公司 An advertising method and device for equipment sales based on artificial intelligence

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8554602B1 (en) * 2009-04-16 2013-10-08 Exelate, Inc. System and method for behavioral segment optimization based on data exchange
JP6097126B2 (en) * 2013-04-10 2017-03-15 株式会社Ｎｔｔドコモ RECOMMENDATION INFORMATION GENERATION DEVICE AND RECOMMENDATION INFORMATION GENERATION METHOD
CN115761599A (en) * 2022-12-21 2023-03-07 合肥工业大学智能制造技术研究院 A video anomaly detection method and system
CN117119143B (en) * 2023-06-07 2024-07-02 青岛尘元科技信息有限公司 Video investigation system, method, equipment and storage medium based on holographic video
CN117391778A (en) * 2023-10-08 2024-01-12 江苏一方月文化科技有限公司 An intelligent analysis system for advertising data based on data analysis
CN120181922A (en) * 2025-03-20 2025-06-20 北京起创科技有限公司 Real-time push system for short drama advertisements based on video feature extraction and multi-dimensional matching

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119046481A (en) * 2024-11-01 2024-11-29 融安云网(北京)技术有限公司 Multimedia video stream management system and method based on artificial intelligence
CN120166238A (en) * 2025-03-20 2025-06-17 南京数海星辰视觉科技有限公司 Video content analysis and post-production optimization method and system based on cloud computing
CN120198178A (en) * 2025-05-26 2025-06-24 长春慧程科技有限公司 An advertising method and device for equipment sales based on artificial intelligence

Also Published As

Publication number Publication date
CN120410629A (en) 2025-08-01

Similar Documents

Publication Publication Date Title
US20230169081A1 (en) Media names matching and normalization
CN100530196C (en) Quick-speed audio advertisement recognition method based on layered matching
US12413646B2 (en) Methods and apparatus to facilitate meter to meter matching for media identification
WO2013185518A1 (en) Detection method for advertisement segment in television live stream
CN102799605A (en) Method and system for monitoring advertisement broadcast
CN118551293A (en) An artificial intelligence data management method
CN107506424A (en) Data analysing method and device
Berrani et al. A non-supervised approach for repeated sequence detection in TV broadcast streams
CN109145109B (en) User group message propagation abnormity analysis method and device based on social network
CN118710338A (en) A marketing plan push system and method based on data analysis
CN120410629B (en) Advertisement structural feature reverse optimization generation system based on artificial intelligence
WO2019219854A1 (en) Device and method for detecting partial matches between a first time varying signal and a second time varying signal
CN116562934A (en) Anti-cheating method, system, electronic device and storage medium for advertising traffic
CN120070066B (en) Multidimensional tax data analysis and prediction method
CN115908080A (en) Carbon emission optimization method and system based on multidimensional data analysis
CN114757694A (en) An advertising video information push system
CN116823349B (en) A method and system for monitoring and managing advertisement delivery based on data collection
CN118195688A (en) Risk identification method, device, electronic device and storage medium
CN108513140B (en) Method for screening repeated advertisement segments in audio and generating wool audio
CN117251747A (en) A method and system for quickly identifying household change relationships for missing files
CN113901261A (en) Abnormal object cluster positioning method and device, electronic equipment and storage medium
Zhou et al. A new approach for noise data detection based on cluster and information entropy
CN118569931B (en) Multi-channel advertisement casting effect real-time monitoring and analyzing method based on big data
CN120724192B (en) Electric power multi-source data slicing processing method and system based on reinforcement learning
CN120894063B (en) New media product user portrait analysis system and method based on media big data

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant