CN107798563B - Internet advertisement effect evaluation method and system based on multi-mode characteristics - Google Patents

Internet advertisement effect evaluation method and system based on multi-mode characteristics Download PDF

Info

Publication number
CN107798563B
CN107798563B CN201711098741.8A CN201711098741A CN107798563B CN 107798563 B CN107798563 B CN 107798563B CN 201711098741 A CN201711098741 A CN 201711098741A CN 107798563 B CN107798563 B CN 107798563B
Authority
CN
China
Prior art keywords
advertisement
user
sequence
browsing
mode
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
CN201711098741.8A
Other languages
Chinese (zh)
Other versions
CN107798563A (en
Inventor
王红
胡晓红
周莹
于晓梅
房有丽
狄瑞彤
孟广婷
刘海燕
宋永强
王露潼
王倩
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.)
Shandong Normal University
Original Assignee
Shandong Normal University
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 Shandong Normal University filed Critical Shandong Normal University
Priority to CN201711098741.8A priority Critical patent/CN107798563B/en
Publication of CN107798563A publication Critical patent/CN107798563A/en
Application granted granted Critical
Publication of CN107798563B publication Critical patent/CN107798563B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06Q30/0245—Surveys
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data

Landscapes

  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Engineering & Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Finance (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses an internet advertisement effect evaluation method and system based on multi-modal characteristics, which comprises the following steps: step (1): collecting various data: acquiring personal basic information and a cognitive mode of a user, multi-modal behavior characteristics generated by the user in the process of browsing internet webpages and information of the memory degree of the user on advertisements; step (2): data analysis and feature extraction: according to the collected data, characteristics are analyzed and screened through characteristic correlation, and characteristic fusion is carried out; and (3): establishing an advertisement memory model: mining the browsing mode of the user, fusing the browsing mode of the user with the characteristics obtained in the step (2), and constructing an advertisement memory model; and (4): and calculating the memory degree of the user to the advertisement by using the established advertisement memory model so as to evaluate the internet advertisement effect. The optimization layout of the pages in the search engine result page, the evaluation of the effect of the network advertisement and the improvement of the user experience are realized.

Description

Internet advertisement effect evaluation method and system based on multi-mode characteristics
Technical Field
The invention relates to the technical field of Internet advertisement effect evaluation, in particular to an Internet advertisement effect evaluation method and system based on multi-mode characteristics.
Background
The click rate, conversion rate and the like are always used as gold standards for evaluating internet advertisements, with the increase of the internet advertisements and the deep understanding of people on the internet advertisements, unless individual advertisements are rich in creativity and attractiveness, a netizen does not click the advertisements blindly, and the netizen may have a certain impression after browsing the advertisements without clicking the advertisements or saving linked websites, even if the netizen often directly visits the websites in the future. The greatest disadvantage of metrics such as click through rate ignores ads that the audience may notice but not pay for a particular action, while conversion rates blend the user's browsing of web pages with the browsing of ads. On average, click rates of less than 1% have not adequately reflected the true effectiveness of web advertisements. Statistics today show that: the average click rate of the network advertisement is reduced from 30% to below 0.5%. Therefore, the currently popular method for measuring the propagation effect of the banner advertisement is not good enough, which drives us to find a new index which can truly reflect the attention degree of the user to the webpage advertisement.
Therefore, for the brand advertisement, the success of the advertisement is not only determined by whether the advertisement is clicked or read and then the commodity is purchased, but more should be expressed by whether the user notices and remembers the commodity, so as to form a brand effect, create a unique and good brand or product image and improve the offline conversion rate in a longer period. Therefore, the memory of the user for the advertisement after the browsing is finished should be an important criterion for measuring the effectiveness of the advertisement. The study on the memory of the user on the advertisement is of great significance. The advent of eye tracking technology provides a way to capture the user's unconscious attention, combining user eye movement behavior characteristics, mouse behavior characteristics, user characteristics, and advertising characteristics,
in summary, in the prior art, an effective solution is not provided for the problem of how to calculate the advertisement quality more objectively and scientifically in the search engine result page, evaluate the advertisement effect, and enhance the user experience.
Disclosure of Invention
In order to solve the problems, the invention provides an internet advertisement effect evaluation method and system based on multi-modal characteristics. The method takes the memory capacity of a user for advertisements as an important new standard for measuring the advertisement effect, obtains the explicit mouse behavior information of the user by embedding Javascript codes in a search engine result page, obtains the implicit eye movement behavior information of the user by using an eye movement tracking mode, analyzes the influence of page layout and user cognition in multi-mode characteristics on the user behavior, and establishes an advertisement memory model by combining a combination mode and a time sequence relation of links in the result page. The optimization layout of the pages in the search engine result page, the evaluation of the effect of the network advertisement and the improvement of the user experience are realized.
In order to achieve the above object, the present invention adopts a first technical solution:
the internet advertisement effect evaluation method based on the multi-modal characteristics comprises the following steps:
step (1): collecting various data: acquiring personal basic information and a cognitive mode of a user, multi-modal behavior characteristics generated by the user in the process of browsing internet webpages and information of the memory degree of the user on advertisements;
step (2): data analysis and feature extraction: according to the collected data, characteristics are analyzed and screened through characteristic correlation, and characteristic fusion is carried out;
and (3): establishing an advertisement memory model: mining the browsing mode of the user, fusing the browsing mode of the user with the characteristics obtained in the step (2), and constructing an advertisement memory model;
and (4): and calculating the memory degree of the user to the advertisement by using the established advertisement memory model so as to evaluate the internet advertisement effect.
In the step (1), the step of collecting each item of data is as follows:
step (1-1): collecting the characteristics of the user, wherein the characteristics of the user comprise: name, age, gender, user cognitive mode; the cognitive mode of the user is obtained through mosaic graphic test;
step (1-2): collecting self-characteristics of the advertisement, wherein the self-characteristics of the advertisement comprise: the size of the advertisement and the location of the advertisement;
step (1-3): collecting eye movement behavior characteristics: the method comprises the steps that an eye movement tracking mode is utilized, and an eye movement instrument is worn to obtain eye movement behavior characteristics of a user in the process that the user browses internet webpages;
step (1-4): collecting mouse behavior characteristics: embedding Javascript codes for recording mouse tracks in the internet webpage to obtain the mouse behavior characteristics of the advertisement clicking area in the process of browsing the webpage by the user;
step (1-5): acquiring advertisement memory degree: and after each webpage is browsed, carrying out advertisement memory test on the user to obtain the advertisement memory degree of the user.
In the step (1-5), the advertisement memory degree is divided into: "A: it is certain to see that "B: it appears that "C: it does not appear to be seen that the "D: definitely not seen ", four levels.
In the step (2), the steps of data analysis and feature extraction are as follows:
step (2-1): primary screening of characteristics: screening the characteristics of significant correlation with the advertisement memory degree at a rated significance level from the characteristics of the user, the advertisement, the eye movement behavior characteristics and the mouse behavior characteristics in a correlation analysis mode;
step (2-2): and (3) characteristic correlation analysis: performing correlation analysis among the features after the screening in the step (2-1) to remove the features with low correlation among the features;
step (2-3): and (3) feature dimensionality reduction: and (3) performing feature fusion on the features processed in the step (2-2), thereby reducing feature dimension and removing data noise.
In the step (3), the step of establishing the advertisement memory model comprises the following steps:
step (3-1): mining a frequent browsing mode: calculating a frequent browsing mode sequence of a user in a search engine result page;
step (3-2): establishing a model: according to the characteristics obtained in the step (2), combining the browsing mode sequence in the step (3-1), and establishing an advertisement memory model by adopting a Random Forest algorithm;
the step (3-1) comprises the following steps:
step (3-1-1): according to the advertisement positionClassifying the data collected in the step (1), arranging the advertisement positions according to the sequence of the time of the collected users entering each link area, wherein the sequence of the retrieval links in each webpage viewed by each user corresponds to a browsing mode sequence Qi；
Step (3-1-2): adding three attributes to all the data obtained in the step (3-1-1) and initializing: adopted length l ═ l (l)1,l2,...,lp) The element support degree S ═ S (S)1,s2,...,sp) Setting a frequency threshold value s; l1＝0,s1＝0；
The support degree of the elements is the frequency of the appearance of any element at the same position in different sequences;
the adopted length is the number of elements of which the support degree is greater than a set frequency threshold value from the first element in any sequence;
step (3-1-3): calculating the support degree s of the first element of the sequencejIf the support of the first element of the sequence is sjIf the frequency is less than the set frequency threshold s, the support s of the first element of the sequence is enabledj0, and eliminating the current sequence;
step (3-1-4): sorting the rest sequences according to the first element values from large to small, and creating a queue G corresponding to each sorted first element1,G2,...,GtEntering the sequences into different queues according to categories, and deleting the first element of each sequence;
step (3-1-5): updating l and S attributes of the sequence, wherein lj+1＝lj+1，sj+1＝s；
Step (3-1-6): repeating steps (3-1-3) and (3-1-4) until the elements in each sequence end with a support equal to 0;
step (3-1-7): calculating a score F for each sequencei＝li*siFrom FiFind the maximum score maxFiThe sequence corresponding to the maximum score is a frequent browsing mode sequence, and a browsing mode sequence Q is outputiOtherwise, judging the sequence as an infrequent browsing mode sequence。
The step (3-2) comprises the following steps:
step (3-2-1): screening out data items containing the frequent browsing pattern sequence in the data set collected in the step (1) according to the mined frequent browsing pattern sequence;
step (3-2-2): according to the characteristics obtained in the step (2-3), calculating the advertisement memory degree with the maximum probability value of all users in each internet webpage by using a Random Forest algorithm;
step (3-2-3): according to the characteristics obtained in the step (2-3), calculating the advertisement memory degree with the maximum probability value of the data item with the frequent browsing mode sequence in each internet webpage by using a Random Forest algorithm;
step (3-2-4): and (3) comparing the advertisement memory degree with the maximum probability value calculated in the step (3-2-2) of the data item of each webpage viewed by the same user with the advertisement memory degree with the maximum probability value calculated in the step (3-2-3), and updating the result obtained in the step (3-2-2) according to the advertisement memory degrees with the large probability values.
Internet advertisement effectiveness evaluation system based on multi-modal characteristics includes: a memory, a processor, and computer instructions stored on the memory and executed on the processor, the computer instructions when executed performing the steps of:
step (1): collecting various data: acquiring personal basic information and a cognitive mode of a user, multi-modal behavior characteristics generated by the user in the process of browsing internet webpages and information of the memory degree of the user on advertisements;
step (2): data analysis and feature extraction: according to the collected data, characteristics are analyzed and screened through characteristic correlation, and characteristic fusion is carried out;
and (3): establishing an advertisement memory model: mining the browsing mode of the user, fusing the browsing mode of the user with the characteristics obtained in the step (2), and constructing an advertisement memory model;
and (4): and calculating the memory degree of the user to the advertisement by using the established advertisement memory model so as to evaluate the internet advertisement effect.
A computer readable storage medium having stored thereon computer instructions which, when executed by a processor, perform the steps of:
step (1): collecting various data: acquiring personal basic information and a cognitive mode of a user, multi-modal behavior characteristics generated by the user in the process of browsing internet webpages and information of the memory degree of the user on advertisements;
step (2): data analysis and feature extraction: according to the collected data, characteristics are analyzed and screened through characteristic correlation, and characteristic fusion is carried out;
and (3): establishing an advertisement memory model: mining the browsing mode of the user, fusing the browsing mode of the user with the characteristics obtained in the step (2), and constructing an advertisement memory model;
and (4): and calculating the memory degree of the user to the advertisement by using the established advertisement memory model so as to evaluate the internet advertisement effect.
The invention has the beneficial effects that:
(1) the internet advertisement effect evaluation system based on the multi-modal characteristics, disclosed by the invention, is used for constructing an advertisement memory model by combining the multi-modal characteristics in a mode of reducing feature dimensions and the like by collecting various types of test behavior information generated by a user in the process of browsing a search engine result page and analyzing the correlation among the characteristics and the correlation between the characteristics and the advertisement memory, and is improved by using a frequent pattern mining method. The layout mode of the advertisements in the search engine result page is optimized, the interaction experience of the user is improved, and the advertisement effect is evaluated more scientifically.
(2) The internet advertisement effect evaluation method based on the multi-mode characteristics firstly provides a new index for measuring the advertisement quality, namely the memory of a user to the advertisement, collects eye movement behavior information, mouse behavior information, advertisement self information and user self information to carry out feature extraction and feature fusion, and combines a user frequent browsing mode to model the memory of the user about the advertisement. The method can accurately predict the memory degree of the advertisement after the user browses the result page, can scientifically and truly reflect the advertisement effect in the search result page, and provides important contribution for the evaluation of the advertisement effect of the internet.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the application and, together with the description, serve to explain the application and are not intended to limit the application.
FIG. 1 is a flow chart of the method for evaluating the effectiveness of Internet advertisement based on multi-modal characteristics according to the present invention;
FIG. 2 is an exemplary illustration of web page material for a search engine results page in a behavioral data collection experiment in accordance with the present invention;
FIG. 3 is a summary of experimental data according to the present invention;
FIG. 4 is a PCA dimension reduction lithograph of eye movement characteristics within a search engine results page in accordance with the present invention;
FIG. 5 is a PCA dimension reduction lithotripsy graph of eye movement characteristics in an advertisement body according to the present invention;
FIG. 6 shows the PCA dimensionality reduction results of the eye movement features of the advertisement of the present invention;
FIG. 7 is a summary of top-5 frequent browsing patterns mined by the present invention;
FIG. 8 is a comparison graph of model accuracy for different methods of construction in the frequent browsing mode according to the present invention;
FIG. 9 is a graph comparing different metrics of the advertisement memory model of the present invention.
Detailed Description
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
Example 1:
in order to solve the problems, the invention provides an internet advertisement effect evaluation method based on multi-modal characteristics. The method takes the memory capacity of a user for advertisements as an important new standard for measuring the advertisement effect, obtains the explicit mouse behavior information of the user by embedding Javascript codes in a search engine result page, obtains the implicit eye movement behavior information of the user by using an eye movement tracking mode, analyzes the influence of page layout and user cognition in multi-mode characteristics on the user behavior, and establishes an advertisement memory model by combining a combination mode and a time sequence relation of links in the result page. The optimization layout of the pages in the search engine result page, the evaluation of the effect of the network advertisement and the improvement of the user experience are realized.
In order to achieve the purpose, the invention adopts the following technical scheme:
an internet advertisement effectiveness evaluation method based on multi-modal characteristics, as shown in fig. 1, includes:
(1) collecting various data: acquiring personal basic information and cognitive style of a user, multi-modal behavior information generated by the user in the browsing process and information of the memory degree of the user on advertisements;
individual characteristics that a person exhibits in cognitive operations. Also known as cognitive style. Cognitive patterns represent the way a person is accustomed to adopting to perceive something outside, and it does not make a good distinction. There are many manifestations of cognitive patterns, such as meditation and impulsivity, flattening and sharpening. The most important of these are the field-dependent and field-independent features proposed by h.a. witness. A person having a field-dependent characteristic tends to view an object as a whole, and is shown to be easily influenced by environmental factors in perception, and a person having a field-independent characteristic tends to receive external stimuli in an analyzed attitude, and is less influenced by environmental factors in perception. The cognitive style is an information processing method that a person prefers to use, and is also called a cognitive method. A cognitive mode comprising: 1. field dependent and field independent; 2. impulse type and sincere type; 3. radial and divergent types; 4. legislative, law enforcement and judicial.
(2) Data analysis and feature extraction: performing correlation analysis to screen features according to the collected multi-modal features, performing normalization processing on data, and performing feature fusion on information sources containing excessive features;
(3) establishing an advertisement memory model: mining a common browsing mode of a user, fusing the self-characteristics of the user, the self-characteristics of the advertisement, the mouse behavior characteristics and the eye tracking characteristics, and constructing an advertisement memory model based on multi-mode characteristics;
(4) evaluating the advertisement memory model: and calculating the memory degree of the advertisement under the model, and evaluating the effectiveness of the established advertisement memory model according to various index values.
In this embodiment, the user cognitive manner is discriminated using a mosaic pattern test proposed by a famous psychologist h.a. witkin et al.
In the present embodiment, the eye movement behavior information is obtained by using an SMI RED (version2.5) eye movement instrument developed by proliter technologies ltd, germany, and the sampling frequency is selected to be 120 Hz.
In this embodiment, the testees are recruited through the web page, all the testees are screened for the visual condition, the naked eye vision or the corrected vision is not less than 1.0, and the eye movement experimenters affected by color blindness, color weakness and the like are removed. 63 tested users were recruited, with a male to female ratio of 1:1.2, ranging in age from 18-21 years, with an average age of 19.7 years.
In this embodiment, the eye tracker for acquiring the eye movement behavior is an SMIRED (version2.5) eye tracker developed by proliter technologies ltd, germany, and the sampling frequency is 120 Hz. The device is non-portable wearing formula equipment, needs to carry out eye movement calibration work before the eye movement experiment begins, and the head can not be removed at will during the data collection of the eye movement by the requirement after the calibration is up to standard.
In this embodiment, the experimental material is divided into two parts, which are the graphic material required for the mosaic test and the web page material required for the user behavior data acquisition. The web page material requires that the search content is as wide as possible in coverage, and the product types are different, for example: amusement parks, ginkgo trees. In order to prevent the impression of network factors and ensure the controllability of variables, the webpage required by the experiment is obtained by capturing, and a Chinese hundredth is selected by the adopted search engine. The obtained webpage only retains ten result links, advertisement links and right-side related recommendations, and other influence links are removed. For convenience of presentation, we divide the page content into different regions of interest and mark them with sequence numbers. Such as: the interest areas of the ten result links are respectively represented as numbers 1-10 from top to bottom, the advertisement interest area is marked as AD, and the right-side related recommendation interest area is marked as R. And Javascript codes for acquiring mouse behavior records are embedded in the webpage. Since it is necessary to study the influence of the advertising position, the experiment was designed to be 3 (page layout) × 2 (cognitive style). In order to prevent fatigue of users, each user executes 6 search tasks at most, namely browsing 6 different webpages, and the users can stop the experiment according to the conditions of the users. An example web page material is shown in figure 2.
In the step (1), the step of collecting each item of data is as follows:
(1-1) a user personal information acquisition module: various items of personal information of the user are collected, such as: the information of the age, the sex and the like of the user, and the cognitive mode of the user is obtained through mosaic graphic test proposed by H.A. Witkin and the like;
(1-2) collecting eye movement behavior information: the method comprises the steps that an eye movement tracking mode is utilized, an eye movement instrument is worn in the process that a user browses a search engine result page to obtain eye movement behavior information of the user, wherein the eye movement behavior information comprises the number of gazing points, gazing duration, review number, entry time and the like;
(1-3) collecting mouse behavior information: embedding Javascript codes for recording mouse actions into a search result page material required by an experiment to obtain mouse behavior information of a click advertisement area in the browsing process of a user, wherein the mouse behavior information comprises a mouse click position, a click number, a moving speed and the like; if the mouse clicking position is in the advertisement area, the user browses the advertisement;
(1-4) acquiring advertisement memory degree: after browsing of each webpage is finished, performing advertisement recognition test to check the memory condition of the tested advertisement.
In the step (1-4), the advertisement memory degree is divided into: "A: it is certain to see that "B: it appears that "C: it does not appear to be seen that the "D: definitely not seen ", four levels.
In this embodiment, the behavior data generated by the user browsing process is 319 pieces in total, and the data summary diagram is shown in fig. 3.
In the step (2), the steps of data analysis and feature extraction are as follows:
(2-1) preliminary screening of characteristics: preliminarily screening characteristics closely related to the advertisement memory degree in other information sources except the information acquired in the step (1-4);
(2-2) feature correlation analysis: performing correlation analysis on the features screened in the step (2-1) to remove correlation features in various information sources;
in this embodiment, four types of feature information are collected, which are: advertisement self information, user information, mouse behavior characteristics, eye movement behavior characteristics. The mouse behavior information is used as explicit information, the eye movement information is used as implicit information, the advertisement information (such as size and position) and the user information (such as cognitive style and advertisement memory). Except that the memory dividing information is used as a class label, the rest information can be used as model characteristics.
In this embodiment, a plurality of characteristics can be obtained by collecting the mouse behavior information, and most of the characteristics have strong positive correlation with the characteristics in the eye movement information. But compared with mouse information, the eye movement information is more scientific, real and instant. Only the click feature with the stronger representativeness is selected. Fig. 4 shows 29 features extracted in the early stage of the experiment, which affect the advertisement memory.
(2-3) feature dimension reduction: and (3) performing feature dimension reduction processing on the information source which still contains a large number of features after the processing of the step (2-2), removing data noise, and extracting components which have important influence on the advertising degree from the multiple features.
In the present embodiment, since the eye movement characteristics are excessive, pca (principal component analysis) dimension reduction operation is performed on the eye movement characteristics. The eye movement features are divided into two sets, and in order to be able to better distinguish the effects of the different sets, the two sets are separated and reduced in dimension. And (3) performing adaptive analysis on the data before dimension reduction, wherein KMO (KaiserMeyer Olkin) is selected, and the test results are all larger than 0.6, which indicates that the experimental data are suitable for PCA dimension reduction. Fig. 5 and 6 are results of PCA dimension reduction of eye movement features of the search engine result page and the advertisement body, where the formulated feature value is greater than 1, and the solid dots in fig. 5 and 6 respectively represent the comprehensive features satisfying the conditions after dimension reduction of the two feature sets. It can be seen that the eye movement feature set of SERPs is subjected to PCA dimension reduction to extract 4 main factors, while the eye movement feature set of advertisement SERPs is subjected to dimension reduction to extract 3 main factors, and the cumulative contributions are 95.4% and 85.6% respectively.
In the step (3), the step of establishing the advertisement memory model comprises the following steps:
(3-1) frequent browsing pattern mining: calculating a Frequent browsing mode of a user in a search engine result page according to an algorithm DFBP (directional frequency browsing patterns);
(3-2) a model building module: according to the characteristics after the characteristic screening and processing, combining with a user frequent browsing mode, and establishing an advertisement memory model by adopting a Random Forest algorithm;
the step (3-1) comprises the following steps:
step (3-1-1): classifying the data collected in the step (1) according to the advertisement positions, wherein each class is arranged according to the sequence of the time when the collected users firstly enter each interest area, and each webpage viewed by each user corresponds to a browsing mode sequence Qi；
For example: an advertising location comprising: the top of the web page, the middle of the web page or the bottom of the web page;
each user views each webpage and corresponds to a browsing mode sequence:
for example, opening the hundred degrees, inputting a search term, wherein the corresponding search term has 10 search links, the number is 1-10, the advertisement area is defined as AD, and the related area of the search term is defined as R;
suppose that the user's eye browsing order is AD, 1,2,3, R, 4, 5, 6, 8, 7, 9, 10, respectively; then the sequence of browsing patterns for the user is: AD → 1 → 2 → 3 → R → 4 → 5 → 6 → 8 → 7 → 9 → 10;
suppose that the user's eye browsing order is 1,2,3, AD, 4, 5, 6, 7, 8, 9, 10, R, respectively; then the sequence of browsing patterns for the user is: 1 → 2 → 3 → AD → 4 → 5 → 6 → 7 → 8 → 9 → 10 → R;
suppose that the user's eye browsing order is AD, 2,3, 4, 5, respectively; then the sequence of browsing patterns for the user is: AD → 2 → 3 → 4 → 5;
step (3-1-2): adding three attributes to all the data obtained in the step (3-1-1) and initializing: adopted length L ═ L1,l2,...,lp) The sequence of support S ═ S (S)1,s2,...,sp) A support threshold s; wherein l1＝0,s1＝0，
The support degree of the elements is recorded as the frequency of the elements, when 300 sequences exist in a data set, the first element is 30 AD sequences, and the support degree of the first element is 30; on the premise that the first elements are consistent, 10 second elements are 1, and the support degree of the second elements is 10; on the premise that the first two elements are identical, the number of the third element is 7, the support degree of the third element is 7, and so on.
The adopted length is specific length information having a frequent condition for a certain sequence.
The sequence calculates the support degree of the first element from the first element, if the support degree of the first element is larger than a support degree threshold value s, then l 11, L ═ 1, proceed, otherwise L10, L ═ 0, end;
when l is1When the support degree of the second element is 1, calculating the support degree of the second element, and if the support degree of the second element is larger than a support degree threshold value s, l2Continue with 2, L (1,2), otherwise keep L 11, and ending when L is not changed to (1);
when l is2When the support degree of the third element is 2, calculating the support degree of the third element, and if the support degree of the third element is greater than a support degree threshold value s, l 33, L ═ 1,2,3), and so onCalculating the subsequent element support degree; otherwise, keeping the state unchanged, and ending.
For example, the browsing pattern sequence AD → 1 → 2 → 3 → R → 4 → 5 → 6 → 8 → 7 → 9 → 10, and the support threshold s is 8. The support degree of the first element AD is 30>8, then l 11, L ═ 1; the second element 1 has a support degree of 10>8, then l 22, L ═ 2; the third element 2 has a support degree of 7<8, then maintain l2And (2) ending after the L is not changed to (1, 2).
And the support degree sequence is used for knowing specific frequency information of a certain sequence. Calculating the support degree of the first element from the first element, and when 300 data exist in a data set and the first element is AD and has 30, calculating the support degree s of the first element130, S ═ 30; on the premise that the first elements are consistent, 10 second elements are 1, and the support degree of the second elements is s210, S ═ 10; on the premise that the first two elements are identical, the number of the third element is 2, and the support degree of the third element is s3＝7，7<S, end, S ═ 30, 10, and so on if the third element meets greater than the support threshold.
A frequency threshold value and a set value. When the support degree of an element is greater than the frequency threshold, the element is considered to be frequent, and s is 8.
Step (3-1-3): calculating the support degree s of the first element of the sequencejIf s isjIf < s, let sequence sj0 and eliminating the sequence;
for example, if the first elements of 30 sequences in the current data set are the same and the first elements are all ADs, the support s of the first elements isj＝30；
Step (3-1-4): sorting the rest sequences according to the first element values from large to small, and creating a queue G corresponding to each sorted first element1,G2,...,GtEntering the sequences into different queues according to categories, and deleting the first element of each sequence;
step (3-1-5): updating the L and S attributes of the sequence, and adding L to L and S respectivelyj+1And sj+1(ii) a Wherein lj+1＝lj+1，sj+1S, original L ═ L1,l2...,lj) And S ═ S1,s2,...,sj) After update, L ═ L1,l2...,lj,lj+1) And S ═ S1,s2,...,sj,sj+1)；
Step (3-1-6): repeating steps (3-1-3) and (3-1-4) until the elements in each sequence end with a support equal to 0;
step (3-1-7): calculating a score F for each sequencei＝li*siFrom FiFind the maximum score maxFiThe sequence corresponding to the maximum score is a frequent browsing mode sequence, and a browsing mode sequence Q is outputiAnd otherwise, judging the sequence to be an infrequent browsing mode sequence.
In the step (3-2), the model building module comprises the following steps:
(3-2-1): and (4) dividing the memory of the advertisement into four grades according to the steps (1-4), and gradually reducing the degree. Four levels are quantized:
Figure BDA0001462876260000101
the characteristics that affect the quantization level are arguments expressed as: x ═ X1,x2,...,xn)。
Screening out a data set subset containing frequent browsing patterns in a data set D according to the matching of the frequent browsing patterns mined by the algorithm
Figure BDA0001462876260000102
(3-2-2): according to the characteristics obtained in the step (2-2) and the step (2-3), calculating the advertisement memory degree with the maximum probability value of all users in each search result page by using a Random Forest algorithm
Figure BDA0001462876260000115
(3-2-3): according toThe characteristics obtained in the step (2-2) and the step (2-3) are used for calculating the advertisement memory degree with the maximum probability value of the data with the frequent browsing mode in each search result page by using a Random Forest algorithm
Figure BDA0001462876260000116
(3-2-4): comparing the calculated memory degree and probability value of the same data item in the step (3-2-2) and the step (3-2-3) under the advertisement memory model
Figure BDA0001462876260000111
If it is not
Figure BDA0001462876260000112
Then
Figure BDA0001462876260000113
Not changing, otherwise
Figure BDA0001462876260000114
In this embodiment, the final results are summarized as shown in fig. 7.
In the step (4), the step of evaluating the advertisement memory model comprises the following steps:
(4-1) evaluation frequent pattern improvement algorithm: comparing the advertisement memory degree predicted by the advertisement memory model with a true value, if the values are the same, adding 1 to the score, if the values are not the same, determining the ratio of the score to the number of data items as the model accuracy, and comparing the model accuracy in various original algorithms and the model algorithm accuracy combined with frequent patterns to prove the superiority of the improved algorithm;
in this embodiment, the accuracy of the memory strength is predicted by using several more classical classification methods among various classification methods. The comparison of the accuracy under different methods is shown in fig. 8. The upper dotted line represents the average accuracy when the classification is done purely using the classical method, and the lower represents the average of the classification accuracy after our improvement.
(4-2) a model evaluation module: and comparing the superiority of the memory model established by various algorithms under different measurement indexes.
In this embodiment, each index uses the arithmetic mean of the various algorithms as a baseline, and FIG. 9 lists the MAE and MSE performance indices for several methods that were extracted. Mae (mean Absolute error) is the average of Absolute errors, which better reflects the actual situation of the predicted value error. MSE (mean Squared error) refers to an expectation value of the square of the difference between a parameter estimation value and a parameter true value, the change degree of data can be evaluated, and the smaller the value of MSE is, the better accuracy of the prediction model in describing experimental data is shown.
It can be seen that: firstly, the memory strength predicted by using Random Forest is the best in accuracy and good in stability, and the accuracy of experimental data described by using the Random Forest is high. For analysis reasons, Random Forest is an algorithm that integrates multiple trees by the idea of ensemble learning, and for an input sample, N trees have N classification results. The random forest integrates all classification voting results, and the category with the most voting times is designated as final output; secondly, no matter which classification method is used, the prediction accuracy rate after the improved strategy proposed by the method is better than the original result, and especially for the situation that the original result is lower than the average value, the accuracy rate after the improvement is increased to a greater extent and is more obvious.
The invention has the beneficial effects that:
(1) the internet advertisement effect evaluation system based on the multi-modal characteristics, disclosed by the invention, is used for constructing an advertisement memory model by combining the multi-modal characteristics in a mode of reducing feature dimensions and the like by collecting various types of test behavior information generated by a user in the process of browsing a search engine result page and analyzing the correlation among the characteristics and the correlation between the characteristics and the advertisement memory, and is improved by using a frequent pattern mining method. The layout mode of the advertisements in the search engine result page is optimized, the interaction experience of the user is improved, and the advertisement effect is evaluated more scientifically.
(2) The internet advertisement effect evaluation method based on the multi-mode characteristics firstly provides a new index for measuring the advertisement quality, namely the memory of a user to the advertisement, collects eye movement behavior information, mouse behavior information, advertisement self information and user self information to carry out feature extraction and feature fusion, and combines a user frequent browsing mode to model the memory of the user about the advertisement. The method can accurately predict the memory degree of the advertisement after the user browses the result page, can scientifically and truly reflect the advertisement effect in the search result page, and provides important contribution for the evaluation of the advertisement effect of the internet.
Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using general purpose computer means, or alternatively, they can be implemented using program code that is executable by computing means, such that they are stored in memory means for execution by the computing means, or they are separately fabricated into individual integrated circuit modules, or multiple modules or steps of them are fabricated into a single integrated circuit module. The present invention is not limited to any specific combination of hardware and software.
The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims (4)

1. Internet advertisement effect evaluation system based on multi-mode characteristics, characterized by includes: a memory, a processor, and computer instructions stored on the memory and executed on the processor, the computer instructions when executed performing the steps of:
step (1): collecting various data: acquiring personal basic information and a cognitive mode of a user, multi-modal behavior characteristics generated by the user in the process of browsing internet webpages and information of the memory degree of the user on advertisements;
collecting the characteristics of the user, wherein the characteristics of the user comprise: name, age, gender, user cognitive mode; the cognitive mode of the user is obtained through mosaic graphic test;
collecting self-characteristics of the advertisement, wherein the self-characteristics of the advertisement comprise: the size of the advertisement and the location of the advertisement;
collecting eye movement behavior characteristics: the method comprises the steps that an eye movement tracking mode is utilized, and an eye movement instrument is worn to obtain eye movement behavior characteristics of a user in the process that the user browses internet webpages;
collecting mouse behavior characteristics: embedding Javascript codes for recording mouse tracks in the internet webpage to obtain the mouse behavior characteristics of the advertisement clicking area in the process of browsing the webpage by the user;
step (2): data analysis and feature extraction: according to the collected data, characteristics are analyzed and screened through characteristic correlation, and characteristic fusion is carried out;
and (3): establishing an advertisement memory model: mining the browsing mode of the user, fusing the browsing mode of the user with the characteristics obtained in the step (2), and constructing an advertisement memory model;
in the step (3), the step of establishing the advertisement memory model comprises the following steps:
step (3-1): mining a frequent browsing mode: calculating a frequent browsing mode sequence of a user in a search engine result page;
the step (3-1) comprises the following steps:
step (3-1-1): classifying the data collected in the step (1) according to the advertisement positions, arranging each advertisement position according to the sequence of the time collected when the user enters each link area, wherein the sequence of each user for viewing the search links in each webpage corresponds to a browsing mode sequence Qi；
Step (3-1-2): adding three attributes to all the data obtained in the step (3-1-1) and initializing: adopted length l ═ l (l)1,l2,...,lp) The element support degree S ═ S (S)1,s2,...,sp) Setting a frequency threshold value s; l1＝0,s10; the support degree of the elements is the frequency of the appearance of any element at the same position in different sequences; the adopted length is the number of elements of which the support degree is greater than a set frequency threshold value from the first element in any sequence;
step (3-1-3): computingSupport s of sequence head elementjIf the support of the first element of the sequence is sjIf the frequency is less than the set frequency threshold s, the support s of the first element of the sequence is enabledj0, and eliminating the current sequence;
step (3-1-4): sorting the rest sequences according to the first element values from large to small, and creating a queue G corresponding to each sorted first element1,G2,...,GtEntering the sequences into different queues according to categories, and deleting the first element of each sequence;
step (3-1-5): updating l and S attributes of the sequence, wherein lj+1＝lj+1，sj+1＝s；
Step (3-1-6): repeating steps (3-1-3) and (3-1-4) until the elements in each sequence end with a support equal to 0;
step (3-1-7): calculating a score F for each sequencei＝li*siFrom FiFind the maximum score maxFiThe sequence corresponding to the maximum score is a frequent browsing mode sequence, and a browsing mode sequence Q is outputiOtherwise, judging the sequence as an infrequent browsing mode sequence;
step (3-2): establishing a model: according to the characteristics obtained in the step (2), combining the browsing mode sequence in the step (3-1), and establishing an advertisement memory model by adopting a Random Forest algorithm;
the step (3-2) comprises the following steps:
step (3-2-1): screening out data items containing the frequent browsing pattern sequence in the data set collected in the step (1) according to the mined frequent browsing pattern sequence;
step (3-2-2): according to the characteristics obtained in the step (2-3), calculating the advertisement memory degree with the maximum probability value of all users in each internet webpage by using a Random Forest algorithm;
step (3-2-3): according to the characteristics obtained in the step (2-3), calculating the advertisement memory degree with the maximum probability value of the data item with the frequent browsing mode sequence in each internet webpage by using a Random Forest algorithm;
step (3-2-4): comparing the advertisement memory degree with the maximum probability value calculated in the step (3-2-2) of the data item of each webpage viewed by the same user with the advertisement memory degree with the maximum probability value calculated in the step (3-2-3), and updating the result obtained in the step (3-2-2) according to the advertisement memory degrees with the large probability values;
and (4): and calculating the memory degree of the user to the advertisement by using the established advertisement memory model so as to evaluate the internet advertisement effect.
2. The system as claimed in claim 1, wherein in the step (2), the steps of data analysis and feature extraction are as follows:
step (2-1): primary screening of characteristics: screening the characteristics of significant correlation with the advertisement memory degree at a rated significance level from the characteristics of the user, the advertisement, the eye movement behavior characteristics and the mouse behavior characteristics in a correlation analysis mode;
step (2-2): and (3) characteristic correlation analysis: performing correlation analysis among the features after the screening in the step (2-1) to remove the features with low correlation among the features;
step (2-3): and (3) feature dimensionality reduction: and (3) performing feature fusion on the features processed in the step (2-2), thereby reducing feature dimension and removing data noise.
3. A computer readable storage medium having stored thereon computer instructions which, when executed by a processor, perform the steps of:
step (1): collecting various data: acquiring personal basic information and a cognitive mode of a user, multi-modal behavior characteristics generated by the user in the process of browsing internet webpages and information of the memory degree of the user on advertisements;
collecting the characteristics of the user, wherein the characteristics of the user comprise: name, age, gender, user cognitive mode; the cognitive mode of the user is obtained through mosaic graphic test;
collecting self-characteristics of the advertisement, wherein the self-characteristics of the advertisement comprise: the size of the advertisement and the location of the advertisement;
collecting eye movement behavior characteristics: the method comprises the steps that an eye movement tracking mode is utilized, and an eye movement instrument is worn to obtain eye movement behavior characteristics of a user in the process that the user browses internet webpages;
collecting mouse behavior characteristics: embedding Javascript codes for recording mouse tracks in the internet webpage to obtain the mouse behavior characteristics of the advertisement clicking area in the process of browsing the webpage by the user;
step (2): data analysis and feature extraction: according to the collected data, characteristics are analyzed and screened through characteristic correlation, and characteristic fusion is carried out;
and (3): establishing an advertisement memory model: mining the browsing mode of the user, fusing the browsing mode of the user with the characteristics obtained in the step (2), and constructing an advertisement memory model;
in the step (3), the step of establishing the advertisement memory model comprises the following steps:
step (3-1): mining a frequent browsing mode: calculating a frequent browsing mode sequence of a user in a search engine result page;
the step (3-1) comprises the following steps:
step (3-1-1): classifying the data collected in the step (1) according to the advertisement positions, arranging each advertisement position according to the sequence of the time collected when the user enters each link area, wherein the sequence of each user for viewing the search links in each webpage corresponds to a browsing mode sequence Qi；
Step (3-1-2): adding three attributes to all the data obtained in the step (3-1-1) and initializing: adopted length l ═ l (l)1,l2,...,lp) The element support degree S ═ S (S)1,s2,...,sp) Setting a frequency threshold value s; l1＝0,s10; the support degree of the elements is the frequency of the appearance of any element at the same position in different sequences; the adopted length is the number of elements of which the support degree is greater than a set frequency threshold value from the first element in any sequence;
step (3-1-3): calculating the support degree s of the first element of the sequencejIf the sequence first elementSupport degree of elements sjIf the frequency is less than the set frequency threshold s, the support s of the first element of the sequence is enabledj0, and eliminating the current sequence;
step (3-1-4): sorting the rest sequences according to the first element values from large to small, and creating a queue G corresponding to each sorted first element1,G2,...,GtEntering the sequences into different queues according to categories, and deleting the first element of each sequence;
step (3-1-5): updating l and S attributes of the sequence, wherein lj+1＝lj+1，sj+1＝s；
Step (3-1-6): repeating steps (3-1-3) and (3-1-4) until the elements in each sequence end with a support equal to 0;
step (3-1-7): calculating a score F for each sequencei＝li*siFrom FiFind the maximum score maxFiThe sequence corresponding to the maximum score is a frequent browsing mode sequence, and a browsing mode sequence Q is outputiOtherwise, judging the sequence as an infrequent browsing mode sequence;
step (3-2): establishing a model: according to the characteristics obtained in the step (2), combining the browsing mode sequence in the step (3-1), and establishing an advertisement memory model by adopting a Random Forest algorithm;
the step (3-2) comprises the following steps:
step (3-2-1): screening out data items containing the frequent browsing pattern sequence in the data set collected in the step (1) according to the mined frequent browsing pattern sequence;
step (3-2-2): according to the characteristics obtained in the step (2-3), calculating the advertisement memory degree with the maximum probability value of all users in each internet webpage by using a Random Forest algorithm;
step (3-2-3): according to the characteristics obtained in the step (2-3), calculating the advertisement memory degree with the maximum probability value of the data item with the frequent browsing mode sequence in each internet webpage by using a Random Forest algorithm;
step (3-2-4): comparing the advertisement memory degree with the maximum probability value calculated in the step (3-2-2) of the data item of each webpage viewed by the same user with the advertisement memory degree with the maximum probability value calculated in the step (3-2-3), and updating the result obtained in the step (3-2-2) according to the advertisement memory degrees with the large probability values;
and (4): and calculating the memory degree of the user to the advertisement by using the established advertisement memory model so as to evaluate the internet advertisement effect.
4. The computer-readable storage medium of claim 3, wherein in the step (2), the step of analyzing data and extracting features comprises:
step (2-1): primary screening of characteristics: screening the characteristics of significant correlation with the advertisement memory degree at a rated significance level from the characteristics of the user, the advertisement, the eye movement behavior characteristics and the mouse behavior characteristics in a correlation analysis mode;
step (2-2): and (3) characteristic correlation analysis: performing correlation analysis among the features after the screening in the step (2-1) to remove the features with low correlation among the features;
step (2-3): and (3) feature dimensionality reduction: and (3) performing feature fusion on the features processed in the step (2-2), thereby reducing feature dimension and removing data noise.
CN201711098741.8A 2017-11-09 2017-11-09 Internet advertisement effect evaluation method and system based on multi-mode characteristics Active CN107798563B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711098741.8A CN107798563B (en) 2017-11-09 2017-11-09 Internet advertisement effect evaluation method and system based on multi-mode characteristics

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711098741.8A CN107798563B (en) 2017-11-09 2017-11-09 Internet advertisement effect evaluation method and system based on multi-mode characteristics

Publications (2)

Publication Number Publication Date
CN107798563A CN107798563A (en) 2018-03-13
CN107798563B true CN107798563B (en) 2020-05-05

Family

ID=61549509

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711098741.8A Active CN107798563B (en) 2017-11-09 2017-11-09 Internet advertisement effect evaluation method and system based on multi-mode characteristics

Country Status (1)

Country Link
CN (1) CN107798563B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110533439A (en) * 2018-05-23 2019-12-03 北京国双科技有限公司 The method and device of value is launched in a kind of evaluation advertisement
CN111221876A (en) * 2020-01-07 2020-06-02 平安科技（深圳）有限公司 Data dimension reduction processing method and device, computer equipment and storage medium
CN111583217A (en) * 2020-04-30 2020-08-25 深圳开立生物医疗科技股份有限公司 Tumor ablation curative effect prediction method, device, equipment and computer medium
CN112181152B (en) * 2020-11-13 2023-05-26 幻蝎科技(武汉)有限公司 Advertisement pushing management method, device and application based on MR (magnetic resonance) glasses
CN115904911A (en) * 2022-12-24 2023-04-04 北京津发科技股份有限公司 Web human factor intelligent online evaluation method, system and device based on cloud server
CN116402553B (en) * 2023-06-07 2023-08-18 江西时刻互动科技股份有限公司 Advertisement effect evaluation method, device and readable storage medium

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005243021A (en) * 2004-02-26 2005-09-08 Microsoft Corp Multimode navigation in graphical user interface computing system
CN101203875A (en) * 2005-03-30 2008-06-18 谷歌公司 Adjusting an advertising cost, such as a per-ad impression cost, using a likelihood that the ad will be sensed or perceived by users
CN102496012A (en) * 2011-12-14 2012-06-13 上海海事大学 Device and method for discovering potential demands based on eye movement tracking and historical behavior
CN103617263A (en) * 2013-11-29 2014-03-05 安徽大学 Automatic TV advertisement movie clip detection method based on multi-mode features
CN104700090A (en) * 2015-03-25 2015-06-10 武汉大学 Method and system for measuring eye movement fixation points based on density
CN106920129A (en) * 2017-03-09 2017-07-04 山东师范大学 A kind of network advertisement effect evaluation system and its method that tracking is moved based on eye

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005243021A (en) * 2004-02-26 2005-09-08 Microsoft Corp Multimode navigation in graphical user interface computing system
CN101203875A (en) * 2005-03-30 2008-06-18 谷歌公司 Adjusting an advertising cost, such as a per-ad impression cost, using a likelihood that the ad will be sensed or perceived by users
CN102496012A (en) * 2011-12-14 2012-06-13 上海海事大学 Device and method for discovering potential demands based on eye movement tracking and historical behavior
CN103617263A (en) * 2013-11-29 2014-03-05 安徽大学 Automatic TV advertisement movie clip detection method based on multi-mode features
CN104700090A (en) * 2015-03-25 2015-06-10 武汉大学 Method and system for measuring eye movement fixation points based on density
CN106920129A (en) * 2017-03-09 2017-07-04 山东师范大学 A kind of network advertisement effect evaluation system and its method that tracking is moved based on eye

Also Published As

Publication number Publication date
CN107798563A (en) 2018-03-13

Similar Documents

Publication Publication Date Title
CN107798563B (en) Internet advertisement effect evaluation method and system based on multi-mode characteristics
CN109145712B (en) Text information fused GIF short video emotion recognition method and system
CN110737783B (en) Method and device for recommending multimedia content and computing equipment
CN111400591B (en) Information recommendation method and device, electronic equipment and storage medium
CN108664932B (en) Learning emotional state identification method based on multi-source information fusion
CN105512687A (en) Emotion classification model training and textual emotion polarity analysis method and system
EP3623998A1 (en) Character recognition
AU2020102036A4 (en) A Method of Clothing Attribute Recognition with Graph Relation Network
CN110705490B (en) Visual emotion recognition method
CN108229999B (en) Method and device for evaluating competitive products
Sultan et al. Sign language identification and recognition: A comparative study
CN108717548B (en) Behavior recognition model updating method and system for dynamic increase of sensors
Li et al. Improved-storygan for sequential images visualization
CN113723530A (en) Intelligent psychological assessment system based on video analysis and electronic psychological sand table
CN111898704A (en) Method and device for clustering content samples
He et al. What catches the eye? Visualizing and understanding deep saliency models
Chen et al. Dynamic graph convolutional network for assembly behavior recognition based on attention mechanism and multi-scale feature fusion
Desai et al. ASL citizen: a community-sourced dataset for advancing isolated sign language recognition
CN107194322A (en) A kind of behavior analysis method in video monitoring scene
CN114758382A (en) Face AU detection model establishing method and application based on adaptive patch learning
Chakraborty et al. Predicting visual attention in graphic design documents
CN109214275B (en) Vulgar picture identification method based on deep learning
Han et al. Human scanpath estimation based on semantic segmentation guided by common eye fixation behaviors
CN111008579A (en) Concentration degree identification method and device and electronic equipment
CN111881007A (en) Operation behavior judgment method, device, equipment and computer readable storage medium

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