A cloud video advertisement intelligent putting management method and system based on information recognition

Abstract

The application discloses a cloud video advertisement intelligent putting management method and system based on information recognition, relates to the technical field of advertisement putting, and comprises the following steps: acquiring user characteristic information; acquiring user visual focus data and behavior data based on data preprocessing according to the user characteristic information; acquiring cloud video advertisement information; and acquiring cloud video advertisement attribute information based on information recognition. Through the eye movement track and user behavior, the behavior semantic knowledge graph constructed based on the multiple source data, the real-time demand and the scenario intention of the user are analyzed in real time, the second gazing area of the user is acquired on the basis of the first gazing area of the user, the user preference label is extracted from the second gazing area of the user to reveal the deep-level preference of the user, the weight of the first preference label is adjusted through the second preference label and the third preference label, the system can be rapidly adapted to the short-term preference change and the popular trend, and the flexibility and timeliness of advertisement putting are enhanced.

Description

Technical Field

[0001] This invention relates to the field of advertising delivery technology, specifically to a cloud video advertising intelligent delivery management method and system based on information recognition. Background Technology

[0002] Cloud video advertising is not only an essential industry need to improve advertising conversion efficiency and optimize the allocation of digital marketing resources by accurately matching user needs, but also a key path to balance commercial value and user experience through technological innovation and promote the transformation of advertising from "traffic delivery" to "value delivery". At the same time, it provides practical support for the integrated application of technologies in multiple fields and the standardized development of the digital ecosystem.

[0003] Current mainstream cloud video advertising systems have significant technical limitations: on the one hand, traditional recommendation methods based on user profiles and basic behavioral data struggle to deeply understand users' real-time needs and contextual intentions during viewing; on the other hand, existing systems generally lack the ability to quickly respond to short-term shifts in user preferences, resulting in a low match between cloud video ad delivery and users' immediate needs, severely impacting the effectiveness of ad delivery. Specifically, these limitations manifest as: (1) a superficial understanding of needs, failing to extract deep semantic information from video content and user interactions; (2) a lagging response mechanism, making it difficult to capture and adapt to dynamic changes in user attention in a timely manner; and (3) insufficient fusion of multimodal data, with biometric data such as eye-tracking trajectories and behaviors not being fully utilized. These problems lead to low ad conversion rates, poor user experience, and loss of commercial value. Summary of the Invention

[0004] To address the aforementioned technical issues, this paper provides a cloud video advertising intelligent delivery management method and system based on information recognition. This technical solution solves the problem that the recommendation methods based on user profiles and basic behavioral data mentioned in the background technology are unable to deeply understand the real-time needs and contextual intentions of users during the viewing process. Existing systems generally lack the ability to quickly respond to short-term shifts in user preferences, resulting in a low degree of matching between cloud video advertising push and users' immediate needs, which seriously affects the delivery effect.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A cloud video advertising intelligent delivery management method based on information recognition includes:

[0007] Obtain user characteristic information, which is the eye movement trajectory and behavioral information recorded when the user watches cloud video advertisements;

[0008] Based on user characteristic information and data preprocessing, user gaze focus data and behavioral data are obtained;

[0009] Obtaining cloud video advertisement information, based on information recognition, obtaining cloud video advertisement attribute information;

[0010] According to the line of sight focus data and the cloud video advertisement attribute information, based on eye tracking and target detection, obtaining the key element information preferred by the user in the cloud video advertisement attribute information;

[0011] According to the key element information preferred by the user and the cloud video advertisement attribute information, based on a deep learning model, obtaining an initial cloud video advertisement, the initial cloud video advertisement being the most matched advertisement in a cloud video advertisement pool obtained by the deep learning model according to the cloud video advertisement attribute information and the key element information preferred by the user;

[0012] Obtaining the eye movement trajectory and behavior information of the user when watching the initial cloud video advertisement, and updating the user's line of sight focus data and behavior data;

[0013] Obtaining multi-source data, constructing a behavior semantic knowledge graph according to the multi-source data, the multi-source data including eye movement trajectory, behavior data, item attribute and behavior semantic information;

[0014] Preferably, according to the line of sight focus data and the cloud video advertisement attribute information, based on eye tracking and target detection, obtaining the key element information preferred by the user in the cloud video advertisement attribute information, specifically comprising:

[0015] According to the cloud video advertisement information, based on picture preprocessing, obtaining a cloud video advertisement picture set, the picture preprocessing including video segmentation and picture noise reduction processing;

[0016] According to the cloud video advertisement picture set, based on information recognition, obtaining cloud video advertisement attribute information, the cloud video advertisement attribute information including item name information contained in the cloud video advertisement, boundary box coordinate information of the region containing the item, and brand information;

[0017] According to the user feature information, based on data preprocessing, obtaining user line of sight focus data, the user line of sight focus data including the fixation duration, the number of times and the vertical distance between the upper eyelid and the lower eyelid of the item region;

[0018] According to the user line of sight focus data, obtaining a user preference index;

[0019] According to the user preference index and the cloud video advertisement attribute information, based on eye tracking and target detection, obtaining the key element information preferred by the user, the key element information including a first key element label and a second key element label.

[0020] Preferably, according to the user line of sight focus data, obtaining the user preference index, specifically comprising:

[0021] According to the user visual line focus data, the article gaze density is obtained;

[0022] The article gaze density is specifically:

[0023] ;

[0024] In the formula, is the gaze density of the i-th article, is the gaze duration on the i-th article, is the total gaze duration of the cloud video advertisement, is the number of gazes on the i-th article for more than 1 second, is the total number of gazes on all articles for more than 1 second;

[0025] According to the vertical distance between the upper eyelid and the lower eyelid, the initial interocular distance of the user and the average interocular distance of the gazed article are obtained;

[0026] According to the initial interocular distance of the user, the average interocular distance of the gazed article, and the gaze density, the user preference index is obtained;

[0027] The user preference index is specifically:

[0028] ;

[0029] In the formula, is the user preference index of the i-th article, which controls the sensitivity to the interocular distance, is the ratio of the average interocular distance of the i-th gazed article to the initial interocular distance of the user.

[0030] Preferably, according to the user preference index and the cloud video advertisement attribute information, the key element information of the user preference is obtained based on eye tracking and target detection, specifically including:

[0031] The bounding box coordinate information corresponding to the top five articles of the user preference index is obtained, and the region within the bounding box is taken as the first region of the user gaze;

[0032] According to the first gaze region of the user, the first key user preference label is obtained, and the first key user preference label is the article name information and brand information within the first gaze region of the user;

[0033] According to the user visual line focus data on the first gaze region of the user, the user deep preference region is obtained based on hierarchical clustering and contour extraction, and the user deep preference region is obtained by dividing the user visual line focus into different clusters based on hierarchical clustering, and then extracting the contour of each cluster to generate a closed boundary region;

[0034] According to the user deep preference area, based on the item gaze density, a user gaze second area is obtained, the user gaze second area being the area with the top five item gaze densities in the user deep preference area;

[0035] According to the user gaze second area, based on target detection, a second key user preference label is obtained,

[0036] The second key user preference label is the item name information in the user gaze second area obtained by target detection.

[0037] Preferably, the cloud video advertisement pool specifically comprises:

[0038] According to the user visual focus data and the behavior data, based on a hybrid expert model, a user preference score is obtained;

[0039] According to the user preference score, based on a feature aggregation method, a user core preference label is obtained;

[0040] According to the cloud video advertisement attribute information and the user core preference label, based on multi-dimensional similarity fusion, a comprehensive matching score of the cloud video advertisement is obtained, and if the comprehensive matching score is greater than a preset threshold, the cloud video advertisement is added to the cloud video advertisement pool;

[0041] According to the user preference score, the advertisement with the lowest user preference score in the cloud video advertisement pool is eliminated;

[0042] The historical user visual focus data is obtained, and according to the historical user visual focus data, a first preference label is obtained, the first preference label being the item name information and brand information corresponding to the item with the largest gaze frequency fluctuation amplitude;

[0043] According to the user preference score and the user core preference label, based on a prediction model, a second preference label is obtained;

[0044] Based on big data analysis, a keyword of a current hot event is obtained as a third preference label;

[0045] The frequency fluctuation amplitude is taken as the weight of the first preference label;

[0046] Based on the third preference label, a hot event keyword is obtained;

[0047] According to the key element information of the user preference and the hot event keyword, the number of times corresponding to the hot event keyword is obtained;

[0048] The ratio of the number of times of the hot event keyword to the number of occurrences of the hot event is taken as a hot value correction coefficient;

[0049] According to the key element information and the second preference label, an intersection label value is obtained, the intersection label value being the total number of the same item name information and brand information in the second preference label as the key element information preferred by the user;

[0050] The heat value correction coefficient is adjusted by the ratio of the intersection label value to the total number of the item name information and brand information contained in the second preference label;

[0051] Based on the heat value correction coefficient, the weight of the first preference label is adjusted to obtain the weight of the first preference label;

[0052] According to the first preference label, the second preference label and the third preference label, the cloud video advertisement pool is updated.

[0053] Preferably, according to the behavior semantic knowledge graph, the cloud video advertisement is adjusted based on a deep learning model, specifically including:

[0054] According to the user preference score, a similar user group is obtained based on a user clustering algorithm, the similar user group being a group whose difference in user preference score for the initial cloud video advertisement is less than 0.1;

[0055] According to the similar user group, a high-value cloud video advertisement set is obtained based on a preference advertisement mining method, the high-value cloud video advertisement set being a set of cloud video advertisements whose user preference score for the same cloud video advertisement is higher than a preset threshold value corresponding to the cloud video advertisement;

[0056] According to the updated user visual focus data, a user real-time gaze intersection is obtained based on eye tracking;

[0057] According to the updated behavior data, a real-time interaction intention vector is obtained based on a behavior coding network;

[0058] According to the user real-time gaze intersection and the real-time interaction intention vector, a real-time preference label is obtained based on a behavior semantic knowledge graph;

[0059] According to the real-time preference label, a cloud video advertisement in the cloud video advertisement pool that meets a screening condition is obtained based on a hierarchical screening method and is added to the high-value cloud video advertisement set, the screening condition being item name information and brand information;

[0060] According to the real-time preference label and the user core preference label, a cloud video advertisement to be delivered is obtained from the high-value cloud video advertisement set based on a deep learning model.

[0061] Further, a cloud video advertisement intelligent delivery management system based on information recognition is proposed, which is used to implement the cloud video advertisement intelligent delivery management method based on information recognition as described above, and includes:

[0062] a main control module, configured to: acquire a group core preference label according to user visual focus point data and behavior data; acquire a high-value cloud video advertisement set based on a preference advertisement mining method according to a similar user group; acquire a user real-time gaze intersection based on eye movement tracking according to the user visual focus point data; acquire a real-time interaction intention vector based on a behavior coding network according to the behavior data; acquire a real-time preference label based on a behavior semantic knowledge graph according to the user real-time gaze intersection and the real-time interaction intention vector; and acquire a cloud video advertisement to be pushed from the high-value cloud video advertisement set based on a deep learning model according to the real-time preference label and the user core preference label;

[0063] an information processing module, configured to: acquire a cloud video advertisement picture set based on picture preprocessing according to cloud video advertisement information; acquire cloud video advertisement attribute information based on information recognition according to the cloud video advertisement picture set; acquire user visual focus point data based on data preprocessing according to user feature information; acquire a user preference index according to the user visual focus point data; and acquire key element information preferred by a user based on eye movement tracking and target detection according to the user preference index and the cloud video advertisement attribute information;

[0064] a cloud video advertisement pool module, configured to: acquire a comprehensive matching score of a new cloud video advertisement according to user visual focus point data and behavior data; if the comprehensive matching score is greater than a preset threshold, add the cloud video advertisement information to a cloud video advertisement pool; eliminate an advertisement with the lowest user preference score in the cloud video advertisement pool according to the user preference score; acquire a first preference label according to historical user visual focus point data; acquire a second preference label according to the user preference score and the user core preference label; acquire a keyword of a current hot event as a third preference label based on big data analysis; adjust a weight of the first preference label according to the second preference label and the third preference label; and update the cloud video advertisement pool according to the first preference label, the second preference label and the third preference label.

[0065] Optionally, the main control module specifically comprises:

[0066] a high-value advertisement unit, configured to: acquire a group core preference label according to user visual focus point data and behavior data; and acquire a high-value cloud video advertisement set based on a preference advertisement mining method according to a similar user group;

[0067] a real-time preference unit, configured to: acquire a user real-time gaze intersection based on eye movement tracking according to user visual focus point data; acquire a real-time interaction intention vector based on a behavior coding network according to behavior data; and acquire a real-time preference label based on a behavior semantic knowledge graph according to the user real-time gaze intersection and the real-time interaction intention vector.

[0068] The advertisement delivery unit is used for obtaining a delivered cloud video advertisement from a high-value cloud video advertisement set based on a deep learning model according to the real-time preference label and the user core preference label.

[0069] Optionally, the information processing module specifically comprises:

[0070] The information acquisition unit is used for acquiring a cloud video advertisement picture set based on picture preprocessing according to cloud video advertisement information, acquiring cloud video advertisement attribute information based on information recognition according to the cloud video advertisement picture set, and acquiring user visual focus data based on data preprocessing according to user feature information.

[0071] The information processing unit is used for acquiring a user preference index according to the user visual focus data, and acquiring key element information of user preference based on eye tracking and target detection according to the user preference index and the cloud video advertisement attribute information.

[0072] Optionally, the cloud video advertisement pool module specifically comprises:

[0073] The updating unit is used for acquiring a comprehensive matching score of a new cloud video advertisement according to the user visual focus data and the behavior data, and adding the cloud video advertisement information into the cloud video advertisement pool if the comprehensive matching score is greater than a preset threshold.

[0074] The elimination unit is used for eliminating an advertisement with the lowest user preference score in the cloud video advertisement pool according to the user preference score.

[0075] The preference unit is used for acquiring a first preference label according to historical user visual focus data, acquiring a second preference label according to the user preference score and the user core preference label, acquiring a keyword of a current hot event as a third preference label based on big data analysis, and adjusting a weight of the first preference label according to the second preference label and the third preference label.

[0076] Compared with the prior art, the present application has the following beneficial effects:

[0077] The application provides a cloud video advertisement intelligent putting management method and system based on information recognition. BRIEF DESCRIPTION OF DRAWINGS

[0078] Figure 1 A cloud video advertisement intelligent putting management method based on information recognition is provided in the application.

[0079] Figure 2 A cloud video advertisement attribute information user preference key element information acquisition flowchart is provided in the application.

[0080] Figure 3 A cloud video advertisement putting flowchart is provided in the application.

[0081] Figure 4 A cloud video advertisement intelligent putting management system structure block diagram based on information recognition is provided in the application. DETAILED DESCRIPTION

[0082] The following description is used to disclose the application so that those skilled in the art can implement the application. The preferred embodiments in the following description are only used as examples, and other obvious modifications can be thought of by those skilled in the art.

[0083] REFERENCE Figure 1 - Figure 3 As shown in the figure, the cloud video advertisement intelligent putting management method based on information recognition in the embodiment of the application comprises:

[0084] User feature information is acquired, and the user feature information is the eye movement track and behavior information recorded when the user watches the cloud video advertisement;

[0085] According to the user feature information, the user visual focus data and behavior data are acquired based on data preprocessing;

[0086] Cloud video advertisement information is acquired, and the cloud video advertisement attribute information is acquired based on information recognition;

[0087] According to the visual focus data and the cloud video advertisement attribute information, the cloud video advertisement attribute information user preference key element information is acquired based on eye movement tracking and target detection;

[0088] Specifically, according to the line-of-sight focus data and the cloud video advertisement attribute information, based on eye tracking and target detection, key element information preferred by the user in the cloud video advertisement attribute information is obtained, specifically including:

[0089] According to the cloud video advertisement information, based on picture preprocessing, a cloud video advertisement picture set is obtained, the picture preprocessing including video segmentation and picture noise reduction processing;

[0090] According to the cloud video advertisement picture set, based on information recognition, cloud video advertisement attribute information is obtained, the cloud video advertisement attribute information including item name information contained in the cloud video advertisement, boundary box coordinate information of an item area, and brand information;

[0091] According to the user feature information, based on data preprocessing, user line-of-sight focus data is obtained, the user line-of-sight focus data including gaze duration, number of times of the item area, and vertical distance between the upper eyelid and the lower eyelid;

[0092] According to the user line-of-sight focus data, a user preference index is obtained;

[0093] According to the user preference index and the cloud video advertisement attribute information, based on eye tracking and target detection, key element information preferred by the user is obtained, the key element information including a first key element label and a second key element label;

[0094] Specifically, according to the user line-of-sight focus data, a user preference index is obtained, specifically including:

[0095] According to the user line-of-sight focus data, item gaze density is obtained;

[0096] The item gaze density is specifically:

[0097] ;

[0098] In the formula, is the gaze density of the i-th item, is the gaze duration on the i-th item, is the total gaze duration of the cloud video advertisement, is the number of times of gaze on the i-th item for more than 1 second, is the total number of times of gaze on all items for more than 1 second;

[0099] According to the vertical distance between the upper eyelid and the lower eyelid, a user initial interocular distance and an average interocular distance of gazed items are obtained;

[0100] According to the user initial interocular distance, the average interocular distance of gazed items, and the gaze density, a user preference index is obtained;

[0101] The user preference index is specifically:

[0102] ;

[0103] wherein, is the user preference index of the i-th item, controlling the sensitivity to the interocular distance, is the ratio of the average interocular distance of the i-th gaze item to the initial interocular distance of the user.

[0104] Specifically, according to the user preference index and the cloud video advertisement attribute information, based on eye tracking and target detection, the key element information of user preference is obtained, specifically including:

[0105] The boundary box coordinate information corresponding to the top five items of the user preference index is obtained, and the area within the boundary box is taken as the first user gaze area;

[0106] According to the first user gaze area, the first key user preference label is obtained, and the first key user preference label is the item name information and brand information in the first user gaze area of the user;

[0107] According to the user visual focus data on the first user gaze area, based on hierarchical clustering and contour extraction, the deep user preference area is obtained, and the deep user preference area is obtained by dividing the user visual focus into different clusters based on hierarchical clustering, and then extracting the contour of each cluster to generate a closed boundary area;

[0108] According to the deep user preference area, based on the item gaze density, the second user gaze area is obtained, and the second user gaze area is the top five areas of the item gaze density in the deep user preference area of the user;

[0109] According to the second user gaze area, based on target detection, the second key user preference label is obtained, and the second key user preference label is the item name information in the second user gaze area obtained by target detection;

[0110] If the target detection obtains the item name information in the second user gaze area, it is taken as the second key user preference label, and if the target detection does not obtain the item name information in the second user gaze area, it is not taken as the second key user preference label;

[0111] In the scheme, the item gaze density is obtained through the user visual line focus data, the user preference index is obtained through the user initial interpupillary distance, the average interpupillary distance of the gazed item and the gaze density, the first area gazed by the user is taken as the user gazed first area according to the top five items corresponding to the boundary box in the user preference index, the user deep preference area is obtained according to the user visual line focus data and the user gazed first area, the user gazed second area is obtained by screening the user deep preference area, and the accuracy and personalized level of advertisement matching are improved according to the data information contained in the user gazed first area and the user gazed second area, thereby providing strong support for realizing intelligent and efficient cloud video advertisement delivery.

[0112] It can be understood that the user gazed first area focuses on the top five items most concerned by the user in the cloud video advertisement, which can improve the accuracy of advertisement matching, but in actual application process, only according to the user gazed first area may ignore important preference points, resulting in that the advertisement recommendation is not comprehensive and refined enough, for example, the user gazed first area is the area where a car is located, but the user is more interested in the tires, rearview mirror and tail light of the car, in this case, only according to the user gazed first area cannot capture the deep preference of the user, and the user gazed second area obtained through the user visual line focus can reveal the deep preference and potential preference of the user, dynamically adapt to the changes of the user, and provide more personalized and accurate advertisement recommendation, and the information of the user gazed first area and the user gazed second area can comprehensively understand the user demand, and improve the effect and relevance of advertisement delivery.

[0113] According to the key element information of the user preference and the cloud video advertisement attribute information, an initial delivered cloud video advertisement is obtained based on a deep learning model, the initial delivered cloud video advertisement being the most matched advertisement in a cloud video advertisement pool obtained by the deep learning model according to the cloud video advertisement attribute information and the key element information of the user preference;

[0114] Specifically, the cloud video advertisement pool specifically includes:

[0115] According to the user visual line focus data and the behavior data, a user preference score is obtained based on a hybrid expert model;

[0116] According to the user preference score, a user core preference label is obtained based on a feature aggregation method;

[0117] According to the cloud video advertisement attribute information and the user core preference label, a comprehensive matching score of a new cloud video advertisement is obtained based on multi-dimensional similarity fusion, and if the comprehensive matching score is greater than a preset threshold, the cloud video advertisement information is added to the cloud video advertisement pool;

[0118] The preset threshold is 0.75.

[0119] Based on user preference scores, ads with the lowest user preference scores in the cloud video ad pool are eliminated;

[0120] Obtain historical user gaze focus data, and based on the historical user gaze focus data, obtain the first preference label, which is the item name information and brand information corresponding to the item with the largest fluctuation in the number of gazes;

[0121] Based on user preference scores and core user preference tags, a second preference tag is obtained using a prediction model;

[0122] Based on big data analysis, keywords of current trending events are obtained as third-preference tags;

[0123] Each 10-minute time window is used as a weight for the first preference label, with the frequency fluctuation amplitude as the weight.

[0124] The weight of the first preference label is as follows:

[0125] ;

[0126] In the formula, The weight of the first preference label, For users in the first The time window for the first Number of times an item is viewed. It is a function with maximum value. It is a minimum value function;

[0127] Based on third-party preference tags, obtain keywords for trending events;

[0128] Based on key element information of user preferences and keywords of trending events, obtain the frequency of corresponding keywords of trending events;

[0129] The ratio of the number of times keywords related to trending events appear to the number of times trending events occur is used as the heat value correction coefficient.

[0130] The calorific value correction coefficient is specifically as follows:

[0131] ;

[0132] In the formula, The number of trending event keywords is obtained based on key element information of user preferences and trending event keywords. This represents the total number of times trending events occurred, as determined by big data analysis.

[0133] Based on the key element information of user preferences and the second preference label, obtain the intersection label value, where the intersection label value is the total number of item name information and brand information that are the same as the key element information of user preferences in the second preference label;

[0134] The heat value correction coefficient is adjusted according to a ratio of the intersection label value to a total number of the item name information and the product name information contained in the second preference label;

[0135] The heat value correction coefficient is adjusted according to a ratio of the intersection label value to a total number of the item name information and the product name information contained in the second preference label;

[0136] ;

[0137] In the formula, is a ratio of the intersection label value to a total number of the item name information and the product name information contained in the second preference label;

[0138] The weight of the first preference label is adjusted based on the heat value correction coefficient to obtain the weight of the first preference label;

[0139] The weight of the first preference label is adjusted based on the heat value correction coefficient to obtain the weight of the first preference label;

[0140] ;

[0141] In the formula, if is greater than 0.8, then is re-assigned as 0.8, and if is less than 0.5, then is re-assigned as 0.5;

[0142] The cloud video advertisement pool is updated according to the first preference label, the second preference label and the third preference label;

[0143] Specifically, the cloud video advertisement pool is updated according to the first preference label, the second preference label and the third preference label, and specifically includes:

[0144] The first preference matching score of the cloud video advertisement is obtained based on multi-dimensional similarity fusion according to the first preference label and the cloud video advertisement attribute;

[0145] The second preference matching score of the cloud video advertisement is obtained based on multi-dimensional similarity fusion according to the second preference label and the cloud video advertisement attribute;

[0146] The third preference matching score of the cloud video advertisement is obtained based on multi-dimensional similarity fusion according to the third preference label and the cloud video advertisement attribute;

[0147] The preference matching score is obtained according to the first preference matching score, the second preference matching score and the third preference matching score;

[0148] The preference matching score is obtained according to the first preference matching score, the second preference matching score and the third preference matching score;

[0149] ;

[0150] wherein, is a first preference matching score, is a second preference matching score, is a third preference matching score;

[0151] If is greater than 0.75, the cloud video advertisement is added to the cloud video advertisement pool;

[0152] The eye movement trajectory and behavior information of the user when watching the initially launched cloud video advertisement are obtained, and the user visual focus data and behavior data are updated;

[0153] In the scheme, the user preference score is obtained through the user visual focus data and behavior data, the first stage of updating of the advertisements in the cloud video advertisement pool is performed according to the comparison of the user preference score with the set threshold, the first preference label is obtained according to the historical user visual focus data, the second preference label is obtained according to the user preference score and the user core preference label, the third preference label is obtained based on big data analysis, the frequency fluctuation amplitude is taken as the weight of the first preference label, the ratio of the number of hot event keywords in the second preference label to the number of hot events is taken as a hot value correction coefficient, the intersection label value is obtained according to the key element information of the user preference and the second preference label, the hot value correction coefficient is adjusted according to the ratio of the intersection label value to the total number of the item name information and the product name information contained in the second preference label, the weight of the first preference label is adjusted according to the adjusted hot value correction coefficient, the second stage of updating of the advertisement pool is performed according to the adjusted first preference label weight, and accurate matching and efficient management of the advertisement content are realized.

[0154] It can be understood that the first preference label, the second preference label and the third preference label can quickly match and cover multi-dimensional preferences, but in the actual process, this method has the problems of deviation caused by static weight allocation and lack of real-time adjustment mechanism, the weight of the first preference label is adjusted only according to the third preference label, and there is the problem of excessive weight adjustment, for example, the user is interested in cars, which happen to be the current hot event, and the first preference label weight adjusted accordingly will be smaller than the actual situation, while according to the second preference label and the third preference label, the first preference label weight will be adjusted upward when facing the above situation, so that it is closer to the real situation, in addition, adjusting the weight of the first preference label according to the second preference label and the third preference label can better adapt to the short-term preference fluctuation of the user and the change of popular trends, and optimize the management of the advertisement pool.

[0155] Multi-source data are obtained, and a behavior semantic knowledge graph is constructed according to the multi-source data, wherein the multi-source data include eye movement trajectory, behavior data, item attribute and behavior semantic information;

[0156] According to the behavior semantic knowledge graph, the cloud video advertisement is adjusted based on a deep learning model.

[0157] Specifically, according to the behavior semantic knowledge graph, the cloud video advertisement is adjusted based on a deep learning model, and specifically includes:

[0158] According to the user preference score, a similar user group is obtained based on a user clustering algorithm, and the similar user group is a group with a difference of less than 0.1 in the user preference score of the initial cloud video advertisement.

[0159] According to the similar user group, a high-value cloud video advertisement set is obtained based on a preference advertisement mining method, and the high-value cloud video advertisement set is a set of cloud video advertisements whose user preference scores of the similar user group are higher than a preset threshold.

[0160] According to the updated user visual focus data, a user real-time gaze intersection is obtained based on eye tracking.

[0161] According to the updated behavior data, a real-time interaction intention vector is obtained based on a behavior coding network.

[0162] According to the user real-time gaze intersection and the real-time interaction intention vector, a real-time preference label is obtained based on the behavior semantic knowledge graph.

[0163] According to the real-time preference label, a cloud video advertisement in a cloud video advertisement pool that meets a screening condition is obtained based on a hierarchical screening method and added to the high-value cloud video advertisement set, and the screening condition is an item name information and a brand information.

[0164] According to the real-time preference label and the user core preference label, a cloud video advertisement for delivery is obtained from the high-value cloud video advertisement set based on a deep learning model.

[0165] In this scheme, similar user groups are mined by introducing a user clustering algorithm, a high-value advertisement set is constructed by combining a preference advertisement mining method, user real-time gaze intersections and interaction intention vectors are further obtained by using eye tracking and a behavior coding network, and real-time preference labels are generated based on a behavior semantic knowledge graph to dynamically capture user preferences. On this basis, the high-value advertisement set is updated by a hierarchical screening method, and the most suitable advertisement is selected from the advertisement set by using a deep learning model in combination with the user core preference label. This method effectively integrates the long-term preference and short-term behavior intention of the user, improves the real-time, precision and personalization level of advertisement matching, and significantly enhances the advertisement delivery effect and user experience.

[0166] Reference Figure 4As shown, further, in combination with the above-mentioned cloud video advertisement intelligent putting management method based on information recognition, a cloud video advertisement intelligent putting management system based on information recognition is proposed, comprising:

[0167] A main control module is configured to obtain a user preference score based on a hybrid expert model according to user visual focus data and behavior data, obtain a similar user group based on a user clustering algorithm according to the user preference score, obtain a group core preference label based on a feature aggregation method according to the preference score of the similar user group, obtain a high-value cloud video advertisement set based on a preference advertisement mining method according to the similar user group, obtain a user real-time gaze intersection based on eye movement tracking according to the user visual focus data, obtain a real-time interaction intention vector based on a behavior coding network according to the behavior data, obtain a real-time preference label based on a behavior semantic knowledge graph according to the user real-time gaze intersection and the real-time interaction intention vector, and obtain a put cloud video advertisement from the high-value cloud video advertisement set based on a deep learning model according to the real-time preference label and the user core preference label.

[0168] An information processing module is configured to obtain a cloud video advertisement picture set based on picture preprocessing according to cloud video advertisement information, obtain cloud video advertisement attribute information based on information recognition according to the cloud video advertisement picture set, obtain user visual focus data based on data preprocessing according to user feature information, obtain a user preference index according to the user visual focus data, and obtain key element information of user preference based on eye movement tracking and target detection according to the user preference index and the cloud video advertisement attribute information.

[0169] The cloud video advertisement pool module is configured to: acquire a user preference score based on a hybrid expert model according to user visual focus data and behavior data; acquire a user core preference label based on a feature aggregation method according to the user preference score; acquire a comprehensive matching score of a new cloud video advertisement based on multi-dimensional similarity fusion according to cloud video advertisement attribute information and the user core preference label; if the comprehensive matching score is greater than a preset threshold, add the cloud video advertisement information to the cloud video advertisement pool; eliminate an advertisement with the lowest user preference score in the cloud video advertisement pool according to the user preference score; acquire historical user visual focus data; acquire a first preference label according to the historical user visual focus data; acquire a second preference label based on a prediction model according to the user preference score and the user core preference label; acquire a keyword of a current hot event as a third preference label based on big data analysis; take a frequency fluctuation amplitude as a weight of the first preference label; acquire a hot event keyword based on the third preference label; acquire a number corresponding to the hot event keyword according to key element information of user preference and the hot event keyword; take a ratio of the number of the hot event keyword to a number of occurrences of the hot event as a hot value correction coefficient; acquire an intersection label value according to the key element information of user preference and the second preference label; adjust the hot value correction coefficient according to a ratio of the intersection label value to a total number of item name information and item information contained in the second preference label; adjust the weight of the first preference label based on the adjusted hot value correction coefficient; and acquire the weight of the first preference label, and update the cloud video advertisement pool according to the first preference label, the second preference label and the third preference label.

[0170] The main control module specifically includes:

[0171] The high-value advertisement unit is configured to: acquire a user preference score based on a hybrid expert model according to user visual focus data and behavior data; acquire a similar user group based on a user clustering algorithm according to the user preference score; acquire a group core preference label based on a feature aggregation method according to a preference score of the similar user group; and acquire a high-value cloud video advertisement set based on a preference advertisement mining method according to the similar user group.

[0172] The real-time preference unit is configured to: acquire a user real-time gaze intersection based on eye movement tracking according to user visual focus data; acquire a real-time interaction intention vector based on a behavior coding network according to behavior data; and acquire a real-time preference label based on a behavior semantic knowledge graph according to the user real-time gaze intersection and the real-time interaction intention vector.

[0173] The advertisement delivery unit is configured to: acquire a delivered cloud video advertisement from the high-value cloud video advertisement set based on a deep learning model according to the real-time preference label and the user core preference label.

[0174] The information processing module specifically comprises:

[0175] The information acquisition unit is configured to acquire a cloud video advertisement picture set based on picture preprocessing according to cloud video advertisement information, acquire cloud video advertisement attribute information based on information recognition according to the cloud video advertisement picture set, and acquire user visual focus data based on data preprocessing according to user feature information.

[0176] The information processing unit is configured to acquire a user preference index according to the user visual focus data, and acquire key element information of user preference based on eye tracking and target detection according to the user preference index and the cloud video advertisement attribute information.

[0177] The cloud video advertisement pool module specifically comprises:

[0178] The updating unit is configured to acquire a user preference score based on a hybrid expert model according to the user visual focus data and behavior data, acquire a user core preference label based on a feature aggregation method according to the user preference score, acquire a comprehensive matching score of a new cloud video advertisement based on multi-dimensional similarity fusion according to the cloud video advertisement attribute information and the user core preference label, and add the cloud video advertisement information to the cloud video advertisement pool if the comprehensive matching score is greater than a preset threshold.

[0179] The elimination unit is configured to eliminate an advertisement with the lowest user preference score in the cloud video advertisement pool according to the user preference score.

[0180] The preference unit is configured to acquire historical user visual focus data, acquire a first preference label according to the historical user visual focus data, acquire a second preference label based on a prediction model according to the user preference score and the user core preference label, acquire a keyword of a current hot event as a third preference label based on big data analysis, take a frequency fluctuation amplitude as a weight of the first preference label, acquire a hot event keyword based on the third preference label, acquire a number corresponding to the hot event keyword according to the key element information of user preference and the hot event keyword, take a ratio of the number of the hot event keyword to the number of occurrences of the hot event as a hot value correction coefficient, acquire an intersection label value according to the key element information of user preference and the second preference label, adjust the hot value correction coefficient according to a ratio of the intersection label value to a total number of item name information and brand information contained in the second preference label, adjust the weight of the first preference label based on the adjusted hot value correction coefficient, and acquire the weight of the first preference label.

[0181] In summary, the application has the advantages that: through eye movement track and user behavior, the behavior semantic knowledge graph constructed based on multi-source data is used to analyze real-time demand and scenario intention of the user in real time, the second gaze area of the user is acquired based on the first gaze area of the user, the user preference label is extracted from the second gaze area of the user to reveal deep preference of the user, the weight of the first preference label is adjusted through the second preference label and the third preference label, so that the system can quickly adapt to short-term preference change and popular trend, and the flexibility and timeliness of advertisement delivery are enhanced.

[0182] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection required by the present application is defined by the appended claims and their equivalents.

Claims

1. A cloud video advertisement intelligent delivery management method based on information recognition, characterized in that, The method comprises the following steps: acquiring user feature information, which is the eye movement track and behavior information recorded when the user watches the cloud video advertisement; acquiring user visual focus data and behavior data based on data preprocessing according to the user feature information; acquiring cloud video advertisement information and acquiring cloud video advertisement attribute information based on information recognition; acquiring key element information preferred by the user in the cloud video advertisement attribute information based on eye movement tracking and target detection according to the visual focus data and the cloud video advertisement attribute information; acquiring the initial cloud video advertisement based on a deep learning model according to the key element information preferred by the user and the cloud video advertisement attribute information, wherein the initial cloud video advertisement is the most matched advertisement in the cloud video advertisement pool matched by the deep learning model according to the cloud video advertisement attribute information and the key element information preferred by the user, the key element information includes a first key element label and a second key element label, the first key element label is the item name information and brand information in the first gaze area of the user, the second key element label is the item name information in the second gaze area of the user acquired by target detection, and the cloud video advertisement attribute information includes the item name information, the boundary box coordinate information of the item area and the brand information included in the cloud video advertisement; updating the user visual focus data and behavior data by acquiring the eye movement track and behavior information of the user when watching the initial cloud video advertisement; acquiring multi-source data, constructing a behavior semantic knowledge graph according to the multi-source data, and the multi-source data includes eye movement track, behavior data, item attribute and behavior semantic information; adjusting the cloud video advertisement based on the deep learning model according to the behavior semantic knowledge graph; the method for updating the user visual focus data and behavior data by acquiring the eye movement track and behavior information of the user when watching the initial cloud video advertisement, specifically comprising: acquiring the user preference score based on the hybrid expert model according to the user visual focus data and behavior data; acquiring the user core preference label based on the feature aggregation method according to the user preference score; acquiring the comprehensive matching score of the cloud video advertisement based on the multi-dimensional similarity fusion according to the cloud video advertisement attribute information and the user core preference label, and if the comprehensive matching score is greater than a preset threshold, the cloud video advertisement is added to the cloud video advertisement pool; eliminating the advertisement with the lowest user preference score in the cloud video advertisement pool according to the user preference score; acquiring the first preference label according to the historical user visual focus data, wherein the first preference label is the item name information and brand information corresponding to the item with the largest fluctuation amplitude of the gaze frequency; acquiring the second preference label based on the prediction model according to the user preference score and the user core preference label; acquiring the keyword of the current hot event as the third preference label based on big data analysis; taking the frequency fluctuation amplitude as the weight of the first preference label every 10 minutes as a time window; acquiring the hot event keyword based on the third preference label; acquiring the frequency corresponding to the hot event keyword according to the key element information preferred by the user and the hot event keyword; The ratio of the number of hot event keywords to the number of hot event occurrences is used as a hot value correction coefficient; According to the key element information and the second preference label, an intersection label value is obtained, the intersection label value being the total number of item name information and brand information in the second preference label that is the same as the key element information preferred by the user; The hot value correction coefficient is adjusted by the ratio of the intersection label value to the total number of item name information and brand information in the second preference label; Based on the hot value correction coefficient, the weight of the first preference label is adjusted to obtain the weight of the first preference label; According to the first preference label and the cloud video advertisement attribute, a first preference matching score of the cloud video advertisement is obtained based on multi-dimensional similarity fusion; According to the second preference label and the cloud video advertisement attribute, a second preference matching score of the cloud video advertisement is obtained based on multi-dimensional similarity fusion; According to the third preference label and the cloud video advertisement attribute, a third preference matching score of the cloud video advertisement is obtained based on multi-dimensional similarity fusion; According to the first preference matching score, the second preference matching score, and the third preference matching score, a preference matching score G is obtained. If greater than 0.75, then add this cloud video ad to the cloud video ad pool; The eye movement trajectory and behavior information of the user when viewing the initially launched cloud video advertisement are obtained, and the user's visual focus data and behavior data are updated. 2.The cloud video advertisement intelligent delivery management method based on information recognition according to claim 1, characterized in that, According to the visual focus data and the cloud video advertisement attribute information, the key element information preferred by the user is obtained based on eye tracking and target detection, specifically including: According to the cloud video advertisement information, a cloud video advertisement picture set is obtained based on picture preprocessing, the picture preprocessing including video segmentation and picture noise reduction processing; According to the cloud video advertisement picture set, cloud video advertisement attribute information is obtained based on information recognition, the cloud video advertisement attribute information including item name information, boundary box coordinate information of an item region, and brand information contained in the cloud video advertisement; According to the user feature information, user visual focus data is obtained based on data preprocessing, the user visual focus data including the gaze duration, number of times, and vertical distance between the upper eyelid and the lower eyelid of the item region; According to the user visual focus data, a user preference index is obtained. According to the user preference index and the cloud video advertisement attribute information, the key element information preferred by the user is obtained based on eye tracking and target detection. 3.The cloud video advertisement intelligent delivery management method based on information recognition according to claim 2, characterized in that, According to the user visual focus data, a user preference index is obtained, specifically including: According to the user visual focus data, an item gaze density is obtained; The item gaze density is specifically as follows: ； In the formula, is the gaze density of the i-th item, is the gaze duration on the i-th item, is the total gaze duration of cloud video ads, is the number of gazes over 1 second on the i-th item, is the total number of gazes over 1 second on all items; According to the vertical distance between the upper eyelid and the lower eyelid, a user initial interocular distance and an average interocular distance of the gazed item are obtained; According to the user initial interocular distance, the average interocular distance of the gazed item, and the gaze density, a user preference index is obtained; The user preference index is specifically as follows: ； wherein is the user preference index for the i-th item, controlling the sensitivity to the interocular distance, is the ratio of the average interocular distance for the i-th fixated item to the user initial interocular distance.

4. The cloud video advertisement intelligent delivery management method based on information recognition according to claim 3, characterized in that, According to the user preference index and the cloud video advertisement attribute information, the key element information preferred by the user is obtained based on eye tracking and target detection, specifically including: The boundary box coordinate information of the top five items corresponding to the user preference index is obtained, and the region within the boundary box is taken as the first region gazed by the user; According to the first gaze area of the user, a first key user preference label is obtained; According to the user's line of sight focus data on the first area, a user deep preference area is obtained based on hierarchical clustering and contour extraction, which is obtained by dividing the user's line of sight focus into different clusters based on hierarchical clustering, and then extracting the contour of each cluster to generate a closed boundary region; According to the user deep preference area, a user gaze second area is obtained based on item gaze density, which is the top five areas in the user deep preference area in terms of item gaze density; According to the user gaze second area, a second key user preference label is obtained based on target detection.

5. The cloud video advertisement intelligent delivery management method based on information recognition according to claim 4, characterized in that, According to the behavior semantic knowledge graph, the cloud video advertisement is adjusted based on a deep learning model, specifically including: According to the user preference score, a similar user group is obtained based on a user clustering algorithm, which is a group whose user preference score for the initial cloud video advertisement is less than 0.1 different from the user; According to the similar user group, a high-value cloud video advertisement set is obtained based on a preference advertisement mining method, which is a set of cloud video advertisements whose user preference scores for the same cloud video advertisement are higher than a preset threshold in the similar user group; According to the updated user line of sight focus data, a user real-time gaze intersection is obtained based on eye tracking; According to the updated behavior data, a real-time interaction intention vector is obtained based on a behavior encoding network; According to the user real-time gaze intersection and the real-time interaction intention vector, a real-time preference label is obtained based on the behavior semantic knowledge graph; According to the real-time preference label, a cloud video advertisement in the cloud video advertisement pool that meets the screening condition is obtained based on a hierarchical screening method and added to the high-value cloud video advertisement set, the screening condition being item name information and brand information; According to the real-time preference label and the user core preference label, a cloud video advertisement for delivery is obtained from the high-value cloud video advertisement set based on a deep learning model.

6. An information-identification-based cloud video advertisement intelligent delivery management system for implementing an information-identification-based cloud video advertisement intelligent delivery management method according to any one of claims 1-5. It includes: The main control module is used to obtain a group core preference label based on user line of sight focus data and behavior data, obtain a high-value cloud video advertisement set based on a preference advertisement mining method according to a similar user group, obtain a user real-time gaze intersection based on eye tracking according to user line of sight focus data, obtain a real-time interaction intention vector based on a behavior encoding network according to behavior data, obtain a real-time preference label based on a behavior semantic knowledge graph according to the user real-time gaze intersection and the real-time interaction intention vector, and obtain a cloud video advertisement for delivery from the high-value cloud video advertisement set based on a deep learning model according to the real-time preference label and the user core preference label; The information processing module is configured to acquire a cloud video advertisement picture set based on picture preprocessing according to cloud video advertisement information, acquire cloud video advertisement attribute information based on information recognition according to the cloud video advertisement picture set, acquire user visual focus point data based on data preprocessing according to user feature information, acquire a user preference index according to the user visual focus point data, and acquire key element information of user preference based on eye tracking and target detection according to the user preference index and the cloud video advertisement attribute information. The cloud video advertisement pool module is configured to acquire a comprehensive matching score of the cloud video advertisement according to the user visual focus point data and behavior data, add the cloud video advertisement information to the cloud video advertisement pool if the comprehensive matching score is greater than a preset threshold, eliminate an advertisement with the lowest user preference score in the cloud video advertisement pool, acquire a first preference label according to historical user visual focus point data, acquire a second preference label according to the user preference score and the user core preference label, acquire a keyword of a current hot event as a third preference label based on big data analysis, adjust a weight of the first preference label according to the second preference label and the third preference label, and update the cloud video advertisement pool according to the first preference label, the second preference label and the third preference label.

7. The cloud video advertisement intelligent delivery management system based on information recognition according to claim 6, characterized in that, The main control module specifically includes: The high-value advertisement unit is configured to acquire a group core preference label according to the user visual focus point data and the behavior data, and acquire a high-value cloud video advertisement set based on a preference advertisement mining method according to a similar user group. The real-time preference unit is configured to acquire a user real-time gaze intersection based on eye tracking according to the user visual focus point data, acquire a real-time interactive intention vector based on a behavior coding network according to the behavior data, and acquire a real-time preference label based on a behavior semantic knowledge graph according to the user real-time gaze intersection and the real-time interactive intention vector. The advertisement delivery unit is configured to acquire a delivered cloud video advertisement from the high-value cloud video advertisement set based on a deep learning model according to the real-time preference label and the user core preference label. 8.The information-identification-based cloud video advertisement intelligent delivery management system according to claim 6, characterized in that, The information processing module specifically includes: The information acquisition unit is configured to acquire a cloud video advertisement picture set based on picture preprocessing according to cloud video advertisement information, acquire cloud video advertisement attribute information based on information recognition according to the cloud video advertisement picture set, and acquire user visual focus point data based on data preprocessing according to user feature information. The information processing unit is configured to acquire a user preference index according to the user visual focus point data, and acquire key element information of user preference based on eye tracking and target detection according to the user preference index and the cloud video advertisement attribute information. 9.The information-identification-based cloud video advertisement intelligent delivery management system according to claim 6, characterized in that, The cloud video advertisement pool module specifically includes: The update unit is configured to acquire a comprehensive matching score of a new cloud video advertisement according to the user visual focus point data and the behavior data, and add the cloud video advertisement information to the cloud video advertisement pool if the comprehensive matching score is greater than a preset threshold. The elimination unit is used for eliminating the advertisement with the lowest user preference score in the cloud video advertisement pool according to the user preference score; The preference unit is used for obtaining a first preference label according to historical user visual focus data, obtaining a second preference label according to the user preference score and a user core preference label, obtaining a keyword of a current hot event as a third preference label based on big data analysis, and adjusting the weight of the first preference label according to the second preference label and the third preference label.

Images