Advertisement delivery data evaluation system and method based on multi-channel information integration

Abstract

The invention discloses an advertisement putting data evaluation system and method based on multi-channel information integration, and relates to the technical field of advertisement putting, and the method comprises the steps: obtaining an activeness index of a user according to the operation data of a historical user before the self-starting of an advertisement; based on the historical marketing data, obtaining the striking degree and the association degree of the advertisement, and analyzing operation data of the user after the advertisement is automatically started under the activeness index, the striking degree and the association degree; obtaining a self-starting evaluation value of the advertisement according to operation data of the user after self-starting of the advertisement, establishing an advertisement self-starting control model, and controlling a self-starting behavior of the advertisement through the advertisement self-starting control model; the self-starting behavior of the advertisement is controlled, so that the possibility that the user actively closes the advertisement can be reduced, the retention time of the advertisement in the page and the click rate of the user on the advertisement are improved, and the use experience of the user is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of advertising placement, and specifically to an advertising placement data evaluation system and method based on multi-channel information integration. Background Art

[0002] With the popularization of the Internet and the progress of technology, web page advertising has become one of the important forms of advertising placement. Web page advertising has the characteristics of wide dissemination range, strong interactivity, and precise placement, so it has received extensive attention and favor from advertisers; it is a common phenomenon to embed advertisements in web pages, and when the user drags the web page to a certain position, the video advertisement starts to play automatically or the picture advertisement starts to switch automatically. This behavior may affect the user experience. When the user does not need the advertisement, every time the web page is dragged to a certain position, the advertisement will start automatically, which seriously affects the user experience and forces the user to manually close the advertisement or the page; therefore, how to control the self-start behavior of advertisements and improve the user experience has become an urgent problem to be solved. Summary of the Invention

[0003] The purpose of the present invention is to provide an advertising placement data evaluation system and method based on multi-channel information integration to solve the problems raised in the prior art.

[0004] To achieve the above purpose, the present invention provides the following technical solution: An advertising placement data evaluation system based on multi-channel information integration, including a page data acquisition module, an advertisement self-start control module, a self-start analysis module, and a data storage module; the output end of the data acquisition module is connected to the input end of the data storage module to obtain page data and the user's operation data on the page; the output end of the data storage module is connected to the input end of the self-start analysis module for storing the user's behavior data on the web page; the output end of the self-start analysis module is connected to the input end of the advertisement self-start control module for analyzing the real-time behavior data of the user on the web page to obtain the self-start evaluation value of the advertisement, and generating a control instruction for the advertisement self-start according to the self-start evaluation value; the advertisement self-start control module controls the start state of the advertisement according to the received control instruction.

[0005] Specifically, the page data acquisition module further includes a space conversion unit, an operation monitoring unit, a permission acquisition unit, and a keyword extraction unit; the space conversion unit is used to convert the page data into a color space; the operation monitoring unit is used to monitor the user's operation behavior on the page; the permission acquisition unit is used to apply for permission from the user; the keyword extraction unit is used to obtain the keywords in the page.

[0006] Specifically, the self-start analysis module further includes an activity analysis unit, an attention analysis unit, a prominence analysis unit, and a self-start evaluation unit; the activity analysis unit obtains the user's activity based on the operation data of the user on the page; the attention analysis unit obtains the user's attention to keywords according to the scrolling direction of the page; the prominence analysis unit obtains the prominence of the advertisement after self-start according to the color features of the page, the color features before the advertisement self-starts, and the color features after the advertisement self-starts; the self-start evaluation unit is used to calculate the self-start evaluation value of the advertisement of the real-time behavior data.

[0007] Specifically, the self-start evaluation unit determines the influence directions of the prominence and the activity on the operation data, and obtains different clusters according to the prominence and the activity; trains an advertisement self-start control model according to the prominence and the relevance in the same cluster.

[0008] To achieve the above object, the present invention provides the following technical solutions: An advertisement placement data evaluation method based on multi-channel information integration, including the following steps: Obtain the user's behavior data on the web page through user authorization, and obtain real-time behavior data and historical behavior data; the historical behavior data includes historical marketing data and historical user feedback data; the marketing data includes advertisement data before and after self-start; the user feedback data includes page data, the operation data of the user before the advertisement self-starts, and the operation data of the user after the advertisement self-starts; Obtain the user's activity index according to the operation data of the historical user before the advertisement self-starts; obtain the prominence and relevance of the advertisement based on the historical marketing data, and analyze the operation data of the user after the advertisement self-starts under the activity index, prominence, and relevance; obtain the self-start evaluation value of the advertisement according to the operation data of the user after the advertisement self-starts, and establish an advertisement self-start control model, and the advertisement self-start control model is used to control the self-start of the advertisement; Obtain the self-start evaluation value of the current advertisement according to the real-time behavior data, and control the self-start behavior of the advertisement through the advertisement self-start control model.

[0009] Specifically, the steps of judging the user's state according to the operation data of the historical user before the advertisement self-starts further include the following steps: Extract the operation features from the operation data of the historical user before the advertisement self-starts, and obtain the user's activity index act according to the user's operation features, act = ∑w i ×p i , where p i represents the i-th operation feature, and w i represents the weight of the i-th operation feature.

[0010] Specifically, analyzing the operation data of users after the advertisement self-starts under the activity index, prominence, and relevance also includes: obtaining the attention degree of keywords from historical page data, which specifically includes the following steps: Obtain the keywords and their position information in the page from the historical page data; obtain the page scrolling direction data and keyword metrics from the operation data of historical users before the advertisement self-starts, and calculate the attention degree of the keywords according to the following formula, att = ∑u j ×e j , where e j represents the j-th keyword metric, u j represents the weight of the j-th keyword, where, u j =u j 0 +v 1 ×u j 1 +…+v n ×u j n , where u j 0 is the metric of the j-th keyword when the mouse does not perform a rollback operation, u j 1 、…、u j n represent the metrics of the j-th keyword when the mouse performs 1 time、…、n times of rollback operations, v 1 、…、v n represent the reward coefficients when the mouse performs 1 time、…、n times of rollback operations, 1 < v 1 <…<v n .

[0011] Specifically, obtaining the prominence of the advertisement based on historical marketing data also includes the following steps: Step 1, obtain the image data of the web page from the historical page data, convert the image data of the web page to the RGB space, and calculate the color feature A1 for each pixel in the image of the web page; obtain the average color feature of the web page according to the color features of each pixel in the image of the web page; Step 2, in the same way as Step 1, obtain the average color feature A2 of the advertisement before self-start based on the advertisement data before self-start, and obtain the prominence BXM of the advertisement before self-start, BXM = A2 / A1; Step 3: Obtain the average color feature of each image frame of the advertisement after self-start based on the advertisement data before self-start, and generate a first saliency sequence and a second saliency sequence according to the average color feature of each image frame. The first saliency is the quotient of the average color feature of each image frame of the advertisement after self-start and the average color feature of the web page, and the second saliency is the quotient of the average color feature of the next image frame of the advertisement after self-start and the average color feature of the previous image frame of the advertisement. Take the maximum value of the first saliency sequence and the second saliency sequence to obtain the saliency AXM of the advertisement after self-start.

[0012] Specifically, obtaining the relevance of the advertisement based on historical marketing data further includes the following steps: Extract the text feature vector group from the advertisement data after self-start, and obtain the relevance of the advertisement according to the extracted text feature vector group and the user's attention to the keywords on the page.

[0013] Specifically, analyzing the operation data of users after the advertisement self-starts under the activity index, saliency, and relevance further includes the following steps: S100: Determine the influence direction of saliency and activity on the operation data. Obtain the activity index, saliency, relevance, and the operation data of users after the advertisement self-starts from the historical behavior data to form an operation feature combination [act, xm, gld, fk], where xm represents saliency, gld represents relevance, and fk is the operation data of users after the advertisement self-starts. According to the operation results of users, assign values to clicking on the advertisement, placing the advertisement, and closing the advertisement respectively to quantify the operation data of users after the advertisement self-starts. Concatenate the activity index and relevance to obtain the input feature [act, gld], and perform unsupervised classification on the input feature, which specifically includes the following steps: Initialize the parameters, and set the neighborhood radius and minimum number of points of the DBSCAN algorithm. Data point marking: Use all the input features obtained in step S100 as the data set. Select a data point from the data set, and determine the number of other data points included within the neighborhood radius of the selected data point. If the number is not less than the minimum number of points, mark the selected data point as a core point; if the number is less than the minimum number of points, but the selected data point is within the neighborhood radius of a core point, mark the selected point as a boundary point; otherwise, mark the selected point as a noise point. Generate clusters. Select an unvisited data point from the dataset and check whether the data point is marked as a core point. If it is marked as a core point, add the core point and all unvisited data points within the neighborhood radius of the core point to a cluster, and then mark these data points in the cluster as visited. For each newly added core point in the cluster, continue to process the data points within its neighborhood radius. If these neighborhood points have not been visited, check whether they are marked as core points. If they are marked as core points, add the data points within their neighborhood radius to the cluster and mark them as visited. Repeat this step until no new core points can expand the cluster. Parameter optimization. For the data points in the cluster, analyze the relationship between the salience xm and the operation data fk of the user after the advertisement auto-starts in the corresponding operation feature combination. Sort [xm fk] in ascending order according to xm, obtain the distribution data of three cases of fk, obtain the monotonicity based on the distribution data of three cases of fk, and adjust the neighborhood radius and the minimum number of points of the DBSCAN algorithm based on the monotonicity. S200. Obtain the operation feature combination [act, xm, gld, fk] in the same cluster, and construct an advertisement auto-start control model in the cluster according to the operation feature combination: y = a × xm + b × gld + c; where a and b are fitting coefficients, c represents the bias, and a, b, and c are solved by the least squares method; y is the advertisement auto-start evaluation value. S300. Obtain real-time behavior data, extract the keywords of the web page from the real-time page data, and obtain the attention of the web page keywords according to the operation data of the user before the advertisement auto-starts; obtain real-time salience, relevance, and activity indicators from the real-time marketing data and real-time user feedback data; determine the cluster to which the real-time behavior data belongs; input the real-time behavior data into the advertisement auto-start control model corresponding to the cluster to obtain the advertisement auto-start evaluation value of the real-time behavior data; if the advertisement auto-start evaluation value is less than k1, control the advertisement not to auto-start; if the auto-start evaluation value is greater than k2, control the advertisement to auto-start; otherwise, keep the current auto-start state of the advertisement without additional auto-start control; where k2 and k1 are the values assigned for the operations of clicking on the advertisement and closing the advertisement.

[0014] Compared with the prior art, the beneficial effects of the present invention are: controlling the auto-start behavior of the advertisement can reduce the possibility that the user actively closes the advertisement, improve the retention time of the advertisement on the page and the click-through rate of the user on the advertisement, and improve the user experience; analyzing the influence direction of the color of the advertisement on the user operation behavior in different situations, obtaining the advertisement auto-start evaluation value, and controlling the auto-start of the advertisement. Description of the Drawings

[0015] Figure 1This is a schematic structural diagram of an advertising placement data evaluation system based on multi-channel information integration according to the present invention. Specific embodiments

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0017] Embodiment: As Figure 1 shown, the present invention provides a technical solution, an advertising placement data evaluation system based on multi-channel information integration, including a page data acquisition module, an advertisement self-start control module, a self-start analysis module, and a data storage module; the output end of the data acquisition module is connected to the input end of the data storage module to obtain page data and the operation data of the user on the page by the user; the output end of the data storage module is connected to the input end of the self-start analysis module for storing the behavior data of the user on the web page; the output end of the self-start analysis module is connected to the input end of the advertisement self-start control module for analyzing the real-time behavior data of the user on the web page to obtain the self-start evaluation value of the advertisement, and generating a control instruction for the advertisement self-start according to the self-start evaluation value; the advertisement self-start control module controls the start state of the advertisement according to the received control instruction.

[0018] The page data acquisition module further includes a space conversion unit, an operation monitoring unit, a permission acquisition unit, and a keyword extraction unit; the space conversion unit is used to convert the page data into a color space; the operation monitoring unit is used to monitor the operation behavior of the user on the page; the permission acquisition unit is used to apply for permissions from the user; the keyword extraction unit is used to obtain the keywords in the page.

[0019] Advertisement data is also a kind of page data, and advertisement data is also obtained through the page data acquisition module; when the advertisement is in the form of a picture, it is directly obtained through the page data acquisition module; when the advertisement is in the form of a video, the page data acquisition module can collect the advertisement image data under different frames.

[0020] The self-start analysis module further includes an activity analysis unit, a attention analysis unit, a prominence analysis unit, and a self-start evaluation unit; the activity analysis unit obtains the activity of the user based on the operation data of the user on the page; the attention analysis unit obtains the attention of the user to the keywords according to the scrolling direction of the page; the prominence analysis unit obtains the prominence of the advertisement after self-start according to the color characteristics of the page, the color characteristics before the advertisement self-starts, and the color characteristics after the advertisement self-starts; the self-start evaluation unit calculates the self-start evaluation value of the advertisement of the real-time behavior data.

[0021] The self - start evaluation unit determines the influence directions of prominence and activity on operation data, obtains different clusters according to prominence and activity; and trains an advertisement self - start control model based on the prominence and relevance in the same cluster.

[0022] Embodiment: The present invention provides a technical solution, an advertisement placement data evaluation method based on multi - channel information integration, including the following steps: Obtain the user's behavior data on the web page through user authorization, and acquire real - time behavior data and historical behavior data; the historical behavior data includes historical marketing data and historical user feedback data; the marketing data includes advertisement data before and after self - start; the user feedback data includes page data, the operation data of the user before advertisement self - start, and the operation data of the user after advertisement self - start; Obtain the activity index of the user according to the operation data of the historical user before advertisement self - start; obtain the prominence and relevance of the advertisement based on the historical marketing data, and analyze the operation data of the user after advertisement self - start under the activity index, prominence, and relevance; obtain the self - start evaluation value of the advertisement according to the operation data of the user after advertisement self - start, and establish an advertisement self - start control model, which is used to control the self - start of the advertisement; Obtain the self - start evaluation value of the current advertisement according to the real - time behavior data, and control the self - start behavior of the advertisement through the advertisement self - start control model.

[0023] Judging the user's status according to the operation data of the historical user before advertisement self - start further includes the following steps: Extract operation features from the operation data of the historical user before advertisement self - start, and obtain the activity index act of the user according to the user's operation features, act = ∑w i ×p i In the formula, p i represents the i - th operation feature, and w i represents the weight of the i - th operation feature.

[0024] The operation features of the user include but are not limited to: the frequency of clicking the mouse on the page, the speed of mouse movement, the speed of page scrolling, and the speed of dragging the page, etc.; the activity index is used to reflect the activity state of the user.

[0025] Analyzing the operation data of the user after advertisement self - start under the activity index, prominence, and relevance further includes: obtaining the attention degree of keywords from the historical page data, specifically including the following steps: Obtain the keywords and the position information of the keywords in the page from the historical page data; obtain the page scrolling direction data and keyword metrics from the historical operation data of the user before the advertisement self-starts, and calculate the attention of the keywords according to the following formula, att = ∑u j ×e j , where e j represents the j-th keyword metric, and u j represents the weight of the j-th keyword. Among them, u j =u j 0 +v 1 ×u j 1 +…+v n ×u j n , where u j 0 is the metric of the j-th keyword when the mouse does not perform a rollback operation, and u j 1 , …, u j n represent the metrics of the j-th keyword when the mouse performs 1 time, …, n times of rollback operations, and v 1 , …, v n represent the reward coefficients for the mouse to perform 1 time, …, n times of rollback operations. 1 < v 1 < … < v n .

[0026] The keyword metrics include but are not limited to: the number of times the keyword is clicked by the mouse, the time the mouse stays on the keyword, etc.; when the user performs a rollback operation and stops the mouse on a keyword that has already passed or re-clicks a keyword that has already passed, it indicates that the user has a certain interest in the keyword. Therefore, a reward coefficient is added to the rollback operation. For the same keyword, the more times of rollback, the greater the reward coefficient.

[0027] Obtaining the prominence of the advertisement based on historical marketing data further includes the following steps: Step 1, obtain the image data of the web page from the historical page data, convert the image data of the web page to the RGB space, and for each pixel in the image of the web page, calculate the color feature A1; obtain the average color feature of the web page according to the color features of each pixel in the image of the web page; Step 2, in the same way as in Step 1, based on the advertisement data before self-starting, obtain the average color feature A2 of the advertisement before self-starting, and obtain the prominence BXM of the advertisement before self-starting, BXM = A2 / A1; Step 3: Obtain the average color feature of each image frame of the advertisement after self-start based on the advertisement data before self-start. Generate a first saliency sequence and a second saliency sequence according to the average color feature of each image frame. Among them, the first saliency is the quotient of the average color feature of each image frame of the advertisement after self-start and the average color feature of the web page, and the second saliency is the quotient of the average color feature of the subsequent image frame of the advertisement after self-start and the average color feature of the previous image frame of the advertisement. Take the maximum value in the first saliency sequence and the second saliency sequence to obtain the saliency AXM of the advertisement after self-start.

[0028] Optionally, convert the image data to the RGB color space, set the color feature as the brightness value. According to the red, green, and blue channel values in the RGB color space, the brightness value of each pixel point can be calculated, and then the brightness value Y of the page can be obtained according to the brightness value of the pixel point, Y = 0.2123R + 0.7152G + 0.0722B, where R, G, and B are the red, green, and blue channel values of the pixel point respectively. The color feature will affect the user's attention result to the advertisement. The greater the difference in color features, the more likely the user is to notice the advertisement. The user's attention may have a positive impact or a negative impact, which depends on the user's state and the content of the advertisement.

[0029] Obtaining the relevance of the advertisement based on historical marketing data further includes the following steps: Extract the text feature vector group from the advertisement data after self-start, and obtain the relevance of the advertisement according to the extracted text feature vector group and the user's attention to the keywords on the page.

[0030] Obtain the keyword feature vector of the page according to the keywords on the page. According to the keyword feature vector, the text feature vector, and the relevance gld of the advertisement, ; where is the attention to the k-th page keyword after linear transformation. After linearly transforming the attention to all page keywords, the sum of all attentions is made to be one, and ; represents the k-th keyword feature vector in the page; represents the k-th text feature vector of the m-th image frame's text feature vector group after automatic jump; z is the number of text feature vector groups, and M is the number of image frames after automatic jump.

[0031] Analyzing the operation data of the user after the advertisement self-starts under the activity index, saliency, and relevance further includes the following steps: S100, determine the direction of influence of conspicuity and activity on operation data, obtain activity index, conspicuity, relevance and user operation data after the advertisement is launched from historical behavior data, and form an operation feature combination [act, xm, gld, fk], where xm represents conspicuity, gld represents relevance, and fk represents user operation data after the advertisement is launched; assign values ​​to click on the advertisement, place the advertisement and close the advertisement according to the user's operation results, and quantify the user's operation data after the advertisement is launched; perform feature splicing of activity index and relevance to obtain input feature [act, gld], and perform unsupervised classification on the input feature; Optionally, click ads, place ads, and close ads are assigned values ​​of 0, 1, and 2, respectively; The specific steps include: Initialize parameters to set the neighborhood radius and minimum number of points of the DBSCAN algorithm; Data point marking: taking all input features obtained in step S100 as a data set, selecting a data point from the data set, determining the number of other data points contained in the neighborhood radius of the selected data point, and if the number is not less than the minimum number of points, marking the selected data point as a core point; if the number is less than the minimum number of points, but the selected data point is within the neighborhood radius of the core point, marking the selected point as a boundary point; otherwise marking the selected point as a noise point; Generate a cluster, select an unvisited data point from the data set, check whether the data point is marked as a core point, if it is marked as a core point, add the core point and all unvisited data points within the radius of the core point's neighborhood to a cluster, and then mark these data points in the cluster as visited; for each newly added core point in the cluster, continue to process the data points within its neighborhood radius, if these neighborhood points have not been visited, check whether they are marked as core points, if they are marked as core points, add the data points within their neighborhood radius to the cluster and mark them as visited; repeat this step until there are no new core points to expand the cluster; Parameter optimization: For the data points in the cluster, analyze the relationship between the conspicuity xm and the user's operation data fk after the ad is launched in the corresponding operation feature combination, sort [xm fk] in ascending order according to xm, obtain the distribution data of the three cases of fk, and obtain the monotonicity based on the distribution data of the three cases of fk. Based on the monotonicity, adjust the neighborhood radius and minimum number of points of the DBSCAN algorithm; The impact of prominence on operating behavior may be a promoting effect or a hindering effect. For example, when the user activity index is large, it indicates that the user is relatively busy. When the content of the advertisement after self-start is not highly relevant to the content of the page, if the prominence of the advertisement after self-start is high at this time, it may affect the user and prompt the user to close the advertisement. In this case, it is not appropriate to start the advertisement automatically. When the user activity index is low, it means that the user is relatively idle. When the content of the advertisement after self-start is highly relevant to the content of the page, if the prominence of the advertisement after self-start is high at this time, the user is more likely to click on the advertisement after noticing it. At this time, prominence plays a promoting role. The purpose of unsupervised classification is to obtain the impact direction of prominence on operating behavior under different combinations of activity indexes and relevance degrees. To improve the classification effect, parameters need to be optimized. A fixed number of [xm fk] can be regarded as a whole, and then the distribution data of the three cases of fk in each whole in the sequence can be obtained, that is, the distribution of 0, 1, and 2, and the frequencies of 0, 1, and 2 appearing in each whole are obtained to get the frequency change of 0, 1, and 2. If, as xm increases, the distribution of 0, 1, and 2 changes in the same direction, for example, the frequency of 0 decreases and the frequencies of 1 and 2 increase, it indicates good monotonicity. On the contrary, if the frequency of 0 increases and the frequencies of 1 and 2 decrease, it also indicates good monotonicity. If the monotonicity is not obvious, it means that the classification effect is poor, and the neighborhood radius and the minimum number of points need to be adjusted.

[0032] S200, obtain the operation feature combination [act, xm, gld, fk] in the same cluster, and construct an advertisement self-start control model in the cluster according to the operation feature combination: y = a × xm + b × gld + c; where a and b are fitting coefficients, c represents the bias, and a, b, and c are solved by the least squares method; y is the self-start evaluation value of the advertisement. Use historical user feedback data to solve the parameters, obtain xm, gld, and fk in the historical data, use xm and gld as inputs and fk as the output, and train to obtain the fitting coefficients and the bias. During the training process, if the fitting value and the true value of the model are in the same category, the error is set to zero, otherwise the error is retained. For example, fk = 0 indicates that the user has closed the self-started advertisement. If the output fitting result of the model is between (0, 1), it means that the model classification is correct, so no error is generated.

[0033] S300, obtain real-time behavior data, extract keywords of the web page from the real-time page data, and obtain the attention degree of the web page keywords based on the operation data of the user before the advertisement self-starts; obtain real-time prominence, relevance, and activity indicators from the real-time marketing data and real-time user feedback data; determine the cluster to which the real-time behavior data belongs; input the real-time behavior data into the advertisement self-start control model corresponding to the cluster to obtain the self-start evaluation value of the advertisement for the real-time behavior data; if the self-start evaluation value of the advertisement is less than k1, then control the advertisement not to self-start; if the self-start evaluation value is greater than k2, then control the advertisement to self-start; otherwise, keep the current self-start state of the advertisement without additional self-start control; where k2 and k1 are the values assigned to the operations of clicking on the advertisement and closing the advertisement.

[0034] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claims involved.

Claims

1. An advertising data evaluation method based on multi-channel information integration, characterized in that: The following steps are involved: Obtain the user's behavior data on the web page through user authorization, and obtain real-time behavior data and historical behavior data; the historical behavior data includes historical marketing data and historical user feedback data; the marketing data includes advertising data before and after the launch; the user feedback data includes page data, user operation data before the advertisement is launched, and operation data after the advertisement is launched; Obtain the user activity index based on the historical user operation data before the ad is launched; obtain the conspicuity and relevance of the ad based on the historical marketing data, and analyze the user operation data after the ad is launched under the activity index, conspicuity and relevance; Obtaining a self-start evaluation value of the advertisement according to the operation data of the user after the advertisement is self-started, and establishing an advertisement self-start control model, wherein the advertisement self-start control model is used to control the self-start of the advertisement; The self-start evaluation value of the current advertisement is obtained according to the real-time behavior data, and the self-start behavior of the advertisement is controlled through the advertisement self-start control model.

2. The advertising data evaluation method based on multi-channel information integration according to claim 1 is characterized in that: The step of judging the user's status based on the historical user's operation data before the advertisement is automatically started also includes the following steps: Extract the operation features from the historical user's operation data before the ad is launched, and obtain the user's activity index act based on the user's operation features, act=∑w i ×p i , where p i represents the i-th operation feature, w i Represents the weight of the i-th operation feature.

3. The advertising data evaluation method based on multi-channel information integration according to claim 2 is characterized in that: The analyzing of the operation data of the user after the advertisement is launched under the activity index, conspicuity and relevance also includes: obtaining the attention of the keyword from the historical page data, specifically including the following steps: Get the keywords and keyword location information in the page from the historical page data; get the page scrolling direction data and keyword index from the historical user operation data before the ad is launched, and calculate the keyword attention according to the following formula: att = ∑u j ×e j , where e j represents the jth keyword index, u j represents the weight of the jth keyword, where u j =u j 0 +v 1 ×u j 1 +…+v n ×u j n , where u j 0 is the index of the jth keyword when the mouse is not rolling back, u j 1 、…、u j n represents the index of the jth keyword when the mouse performs 1, ..., n rollback operations, v 1 ,…,v n Represents the reward coefficient for the mouse to perform 1, ..., n rollback operations, 1<v 1 <…<v n .

4. The advertising data evaluation method based on multi-channel information integration according to claim 3 is characterized in that: The method of obtaining the conspicuity of the advertisement based on the historical marketing data further comprises the following steps: Step 1: Obtain image data of a web page from historical page data, convert the image data of the web page into RGB space, and calculate a color feature A1 for each pixel in the image of the web page; and obtain an average color feature of the web page based on the color feature of each pixel in the image of the web page; Step 2: In the same manner as step 1, based on the advertisement data before the self-launch, obtain the average color feature A2 of the advertisement before the self-launch, and obtain the conspicuity BXM of the advertisement before the self-launch, where BXM=A2 / A1; Step three, based on the advertising data before self-startup, obtain the average color features of each image frame of the advertisement after self-startup, and generate a first conspicuousness sequence and a second conspicuousness sequence according to the average color features of each image frame, wherein the first conspicuousness is the quotient of the average color features of each image frame of the advertisement after self-startup and the average color features of the web page, and the second conspicuousness is the quotient of the average color features of the image frame after the advertisement after self-startup and the average color features of the image frame before the advertisement; take the maximum value of the first conspicuousness sequence and the second conspicuousness sequence to obtain the conspicuousness AXM of the advertisement after self-startup.

5. The advertising data evaluation method based on multi-channel information integration according to claim 4 is characterized in that: The step of obtaining the relevance of advertisements based on historical marketing data further includes the following steps: A text feature vector group is extracted from the advertisement data after the launch, and the advertisement relevance is obtained according to the extracted text feature vector group and the user's attention to the keywords in the page.

6. The advertising data evaluation method based on multi-channel information integration according to claim 5 is characterized in that: The analysis of the user's operation data after the advertisement is launched under the activity index, conspicuity and relevance also includes the following steps: S100, determine the direction of influence of conspicuity and activity on operation data, obtain activity index, conspicuity, relevance and user operation data after the advertisement is launched from historical behavior data, and form an operation feature combination [act, xm, gld, fk], where xm represents conspicuity, gld represents relevance, and fk represents user operation data after the advertisement is launched; assign values ​​to click on the advertisement, place the advertisement and close the advertisement according to the user's operation results, and quantify the user's operation data after the advertisement is launched; perform feature splicing on the activity index and relevance to obtain input features [act, gld], and perform unsupervised classification on the input features, specifically including the following steps: Initialize parameters to set the neighborhood radius and minimum number of points of the DBSCAN algorithm; Data point marking: taking all input features obtained in step S100 as a data set, selecting a data point from the data set, determining the number of other data points contained in the neighborhood radius of the selected data point, and if the number is not less than the minimum number of points, marking the selected data point as a core point; if the number is less than the minimum number of points, but the selected data point is within the neighborhood radius of the core point, marking the selected point as a boundary point; otherwise marking the selected point as a noise point; Generate a cluster, select an unvisited data point from the data set, check whether the data point is marked as a core point, if it is marked as a core point, add the core point and all unvisited data points within the radius of the core point's neighborhood to a cluster, and then mark these data points in the cluster as visited; for each newly added core point in the cluster, continue to process the data points within its neighborhood radius, if these neighborhood points have not been visited, check whether they are marked as core points, if they are marked as core points, add the data points within their neighborhood radius to the cluster and mark them as visited; repeat this step until there are no new core points to expand the cluster; Parameter optimization: For the data points in the cluster, analyze the relationship between the conspicuity xm and the user's operation data fk after the ad is launched in the corresponding operation feature combination, sort [xm fk] in ascending order according to xm, obtain the distribution data of the three cases of fk, and obtain the monotonicity based on the distribution data of the three cases of fk. Based on the monotonicity, adjust the neighborhood radius and minimum number of points of the DBSCAN algorithm; S200, obtain the operation feature combination [act, xm, gld, fk] in the same cluster, and construct the advertising self-start control model in the cluster according to the operation feature combination: y=a×xm+b×gld+c; where a and b are fitting coefficients, c represents the bias, a, b and c are solved by the least square method; y is the self-start evaluation value of the advertisement; S300, obtain real-time behavior data, extract the keywords of the web page from the real-time page data, and obtain the attention of the web page keywords based on the user's operation data before the advertisement self-starts; obtain real-time eye-catching, relevance and activity indicators from real-time marketing data and real-time user feedback data; determine the cluster to which the real-time behavior data belongs; input the real-time behavior data into the advertisement self-start control model corresponding to the cluster to obtain the advertisement self-start evaluation value of the real-time behavior data; if the advertisement self-start evaluation value is less than k1, control the advertisement not to self-start; if the self-start evaluation value is greater than k2, control the advertisement to self-start; otherwise, maintain the current self-start state of the advertisement without additional self-start control; where k2 and k1 are the values ​​assigned to the operations of clicking on the advertisement and closing the advertisement.

7. An advertising data evaluation system based on multi-channel information integration, characterized in that: It includes a page data acquisition module, an advertisement self-starting control module, a self-starting analysis module and a data storage module; the output end of the data acquisition module is connected to the input end of the data storage module, and the user obtains the page data and the user's operation data on the page; The output end of the data storage module is connected to the input end of the self-start analysis module, and is used to store the user's behavior data on the web page; the output end of the self-start analysis module is connected to the input end of the advertisement self-start control module, and is used to analyze the user's real-time behavior data on the web page to obtain the advertisement's self-start evaluation value, and generate a control instruction for the advertisement's self-start according to the self-start evaluation value; the advertisement self-start control module controls the startup state of the advertisement according to the received control instruction.

8. The advertising data evaluation system based on multi-channel information integration according to claim 7, characterized in that: The page data acquisition module also includes a space conversion unit, an operation monitoring unit, a permission acquisition unit and a keyword extraction unit; the space conversion unit is used to convert page data into a color space; the operation monitoring unit is used to monitor the user's operation behavior on the page; the permission acquisition unit is used to apply for permission from the user; the keyword extraction unit is used to obtain keywords in the page.

9. The advertising data evaluation system based on multi-channel information integration according to claim 8, characterized in that: The self-start analysis module also includes an activity analysis unit, an attention analysis unit, a conspicuousness analysis unit and a self-start evaluation unit; the activity analysis unit obtains the user's activity based on the user's operation data on the page; The attention analysis unit obtains the user's attention to the keyword according to the scrolling direction of the page; the conspicuousness analysis unit obtains the conspicuousness of the advertisement after self-starting according to the color characteristics of the page, the color characteristics of the advertisement before self-starting, and the color characteristics of the advertisement after self-starting; the self-starting evaluation unit is used to calculate the self-starting evaluation value of the advertisement of the real-time behavior data.

10. The advertising data evaluation system based on multi-channel information integration according to claim 9, characterized in that: The self-start evaluation unit determines the direction of influence of the conspicuousness and activity on the operation data, obtains different clusters according to the conspicuousness and activity; and trains the advertisement self-start control model according to the conspicuousness and relevance in the same cluster.

Images