CN-121352884-B - OTT advertisement optimizing system based on pointing behavior of remote controller

CN121352884BCN 121352884 BCN121352884 BCN 121352884BCN-121352884-B

Abstract

The invention relates to the technical field of internet television services, in particular to an OTT advertisement optimizing system based on pointing behaviors of a remote controller, which comprises a data acquisition module, a semantic analysis module and an instruction generation module, wherein the data acquisition module adopts a track processing algorithm based on speed and space-time clustering to distinguish cruising states, staring states and shaking states and extract user intention tracks, the semantic analysis module identifies advertisement intention elements through means such as first frame analysis tracking and the like to generate advertisement semantic element map data, the instruction generation module carries out space-time association mapping on the user intention tracks and the semantic map, calculates element attention scores including high-cost activation times and staring jog amplitude, finally generates intention optimizing instructions and automatically adjusts the layout of the advertisement intention elements. The invention realizes the closed-loop optimization of the OTT advertisement, which is fine, quantifiable and automatic.

Inventors

LAI YILING
ZHANG CHUNYAN
ZHU RONGKUN

Assignees

杭州华数智屏信息技术有限公司

Dates

Publication Date: 20260512
Application Date: 20251216

Claims (7)

1. OTT advertisement optimizing system based on remote controller directional behavior, characterized by comprising: The data acquisition module is used for acquiring an original track data stream acquired by the directional remote controller, adopting track processing to the original track data stream, extracting dormant state data, staring state track data and cruising state track data and generating user intention track data; The semantic analysis module is used for acquiring advertisement visual content data, carrying out semantic understanding on the advertisement visual content through first frame analysis and tracking, identifying advertisement creative elements and generating advertisement semantic element map data containing the advertisement creative elements and corresponding position coordinates; The instruction generation module is used for carrying out association mapping of space and time dimensions on user intention track data and advertisement semantic element map data, calculating attention indexes of advertisement creative elements based on the association mapping, taking predicted advertisement click rate as a main task, predicting sparse signal occurrence probability in the attention indexes as an auxiliary task, jointly optimizing the main task and the auxiliary task, extracting weight learned by the main task, generating element attention scores by weighting and combining the weight and the attention indexes, generating creative optimization instructions based on the element attention scores, and automatically adjusting layout of the advertisement creative elements, wherein the sparse signals comprise high-cost activation times and gaze jog amplitude; The method comprises the steps of calculating attention scores of elements, acquiring attention indexes aiming at advertisement creative elements, namely accumulating total millisecond numbers of staring events related to the elements to reflect time consumption degree of users on the elements, firstly finding out speed, namely recording millisecond numbers spent from advertisement starting to staring of the elements for the first time to reflect the striking degree of the elements, counting the times of event pointed by the remote controller by the users from the dormant state of the remote controller for the first time to reflect the attractive degree of the elements, and counting microscopic intention data of staring jog amplitude on the elements to reflect the staring stability and the concentration degree of the users.
2. The OTT advertisement optimizing system based on the pointing behavior of the remote controller according to claim 1 is characterized in that track processing specifically comprises the steps of receiving an original track data stream, judging cruising track data when the pointer speed of the remote controller is higher than a preset speed threshold, acquiring a coordinate point set of the original track data stream in a preset time window when the pointer speed is not higher than the preset speed threshold, judging staring track data when the spatial variance of coordinate points in spatial distribution is smaller than a clustering threshold, judging jitter track data when the spatial variance of coordinate points in spatial distribution is larger than the clustering threshold, extracting the spatial variance of the staring track data as microscopic intention data, and carrying out smooth processing on the identified staring track data and cruising track data by using a Kalman filtering algorithm to generate user intention track data; The dormant state indicates that the remote controller is static and has no motion within a preset time period, the shaking state indicates unintentional shaking noise data of the hands of the user, the cruising state indicates that the pointer of the remote controller moves from the point A to the point B, and the staring state indicates that the user stares at a certain advertising element.
3. The OTT advertisement optimizing system based on remote control pointing behavior of claim 1, wherein the advertisement semantic element map data comprises main body elements in advertisement materials selected by an advertisement main body frame and outputting categories and boundary frame coordinates, promotion texts, a pronouncing person, brand identification and action call button identification, wherein the promotion texts comprise advertisement text information, identification of person elements in advertisements, comparison and automatic labeling of preset star material library, and brand identification comprises trademark identification and action call button identification in advertisements.
4. The OTT advertisement optimization system based on remote control pointing behavior according to claim 1, wherein when the advertisement visual content data is video, first frame analysis tracking is adopted, and the specific implementation steps include: the method comprises the steps of dividing a video into shots through a scene boundary detection algorithm, identifying all initial advertising creative elements in each shot on the first key frame of each shot, initializing a target tracker for each initial advertising creative element, running the target tracker in the subsequent frames of the shots, predicting and updating the continuously changing bounding box positions of the elements, periodically re-identifying the advertising creative elements, calibrating the position deviation of the tracker and finding new advertising creative elements.
5. The OTT advertisement optimization system based on remote control pointing behavior of claim 1, wherein the implementation of the association map of space and time dimension includes traversing each gaze state event in the user intention track data and acquiring a time stamp and coordinates thereof, the time dimension map including searching advertisement semantic element map data using the time stamp, retrieving advertisement creative elements in corresponding time segments as candidate elements, the space dimension map including acquiring position coordinates of the candidate elements corresponding to the time stamp, and the association decision including attributing attention contribution of the gaze state event to the advertisement creative elements when the coordinates of the gaze state event fall within a bounding box defined by the position coordinates of the candidate elements; The association judging step further comprises the steps that before whether coordinates fall into a boundary box or not is checked, a system firstly dynamically generates an attractive force range for the position coordinates of the candidate elements, wherein the attractive force range is formed by expanding four sides of the candidate elements by a preset pixel value on the basis of an original boundary box, and finally, association judging is performed to check whether the coordinates of the staring state event fall into the expanded attractive force range or not.
6. The OTT advertisement optimization system based on remote control pointing behavior according to claim 1, wherein the element attention score calculation implementation process includes weighting and combining total gaze duration, first discovery speed, high cost activation times and inverse gaze jiggle amplitude by applying an independent weight value respectively; the method comprises the steps of carrying out weight optimization on each weight value through a multi-task learning architecture, constructing a shared representation layer, receiving four indexes as input, learning combined high-dimensional characteristic representation, parallelly constructing a plurality of independent output layers above the shared representation layer, wherein a main task head predicts click rates corresponding to advertisement exposure through a logistic regression layer based on the shared representation, an auxiliary task head predicts occurrence probability of sparse signals based on the same shared representation, jointly optimizes loss functions of all task heads and trains the whole model, and finally takes the weight value output by the main task head as contribution degree of each index to the advertisement click rate.
7. The OTT advertisement optimization system based on remote control pointing behavior of claim 1, wherein the creative optimization instructions specifically comprise element level selection instructions for automatically selecting the highest scoring element among a plurality of alternative advertising creative elements for combined delivery based on the element attention scores, layout dynamic adjustment instructions for automatically adjusting key advertising creative elements to regions determined by the historical intent trace thermodynamic diagram data that have a user attention hover frequency above a preset frequency threshold based on the element attention scores, and automated a/B test instructions for automatically generating advertisement visual content optimized based on element attention scores and automatically conducting a flow a/B test with original advertising visual content.

Description

OTT advertisement optimizing system based on pointing behavior of remote controller Technical Field The invention relates to the technical field of internet television services, in particular to an OTT advertisement optimizing system based on pointing behaviors of a remote controller. Background In the field of large-screen television (OTT) advertising, optimization of creatives is a key element in improving commercial effectiveness of advertisements. Currently, the mainstream advertisement optimization mode in the industry has serious defects. First, depending on the traditional manual a/B test, it has long configuration, execution and analysis cycles, resulting in inefficiency and hysteresis. Second, it relies on ambiguous post-click data. This data is coarse-grained, it can only reflect whether the user has made a final click, but cannot explain which element of the ad creative specifically attracted the user, and thus cannot guide the advertiser through refined creative iterations. In the field of advertisement analysis of PC/mobile internet, there has been a technology for analyzing advertisement layout and optimizing user experience by using a mouse movement trace and a hover thermodynamic diagram. This approach has proven to be effective at the PC side because it can provide more procedural data than "click rate". However, when attempting to apply such a sophisticated advertisement analysis model directly to OTT large screen scenes, an essential obstacle is encountered in that the mouse interaction at the PC side is accurate and its trajectory data can better reflect the user's intention. But OTT interactions are "pitch-back" and users operate in the air using directional remote controllers, whose behavioral data necessarily mixes with a lot of "unintentional" noise from natural hand jitter. Thus, a commercial challenge facing the art is that the "track-based advertising attention attribution" approach, which is mature at the PC end, fails entirely due to the "high noise" nature of OTT remote control data. If the "unintentional" and "conscious" gaze cannot be effectively separated from the noisy trajectory data, the attention signal cannot be accurately attributed to the specific ad creative element. In summary, how to implement an automated, refined and quantifiable OTT ad creative optimization system to replace the inefficient manual a/B test and fuzzy click data is a technical problem to be solved in the art. Therefore, an OTT advertisement optimizing system based on pointing behaviors of a remote controller is provided. Disclosure of Invention The invention aims to provide an OTT advertisement optimizing system based on pointing behaviors of a remote controller, which comprises a data acquisition module, a semantic analysis module, an instruction generation module and an instruction generation module, wherein the data acquisition module adopts a track processing algorithm based on speed and space-time clustering to distinguish a cruising state, a staring state and a dithering state and extract user intention tracks, the semantic analysis module identifies advertisement intention elements through means of first frame analysis and tracking and the like to generate advertisement semantic element map data, the instruction generation module carries out space-time association mapping on the user intention tracks and the semantic map, calculates element attention scores including high cost activation times and staring jog amplitude, finally generates intention optimizing instructions and automatically adjusts the layout of the advertisement intention elements. The invention realizes the closed-loop optimization of the OTT advertisement, which is fine, quantifiable and automatic. In order to achieve the above purpose, the present invention provides the following technical solutions: An OTT advertisement optimization system based on remote control pointing behavior, comprising: The data acquisition module is used for receiving an original track data stream acquired by the directional remote controller, extracting dormant state data, staring state track data and cruising state track data by adopting a track processing algorithm based on speed and space-time clustering on the original track data stream, and distinguishing staring state and shaking state data by calculating spatial variance; The semantic analysis module is used for receiving the advertisement visual content data, carrying out semantic understanding on the advertisement visual content data by applying an image recognition model, recognizing advertisement creative elements in the advertisement visual content and generating advertisement semantic element map data containing the advertisement creative elements and corresponding position coordinates thereof; the instruction generation module is used for carrying out association mapping of space and time dimensions on the user intention track data and the advertisement semantic element map data, c