Big Data Moscow 2018

 

Mark Popov

HCK, Russia

BIO

Expert in machine learning and neural networks. Author and lecturer of “Artificial Intelligence in Media Planning” course for the master program “Communications based on data” of the Higher School of Economics. At the moment, the head of the NSC unit of processing and analysis of big data. More than 15 years in IT, as a developer and leader.
The NSC is the ideological and functional successor in the field of technology and analytics, as well as regional and Internet sales, of Russia’s largest sellers’ structures.

TOPIC

Analysis of the effectiveness of advertising videos using deep learning methods and large data

In the advertising industry, videos placed on sites and ad networks are the most popular format among advertisers. Popularity is due to the high efficiency of the impact of video messages – usually CTR (the ratio of impressions to clicks) of a video is several times higher than of static banners.
This work leads to a deeper understanding of the success of advertising and making decisions how to improve it.
We choose state of the art approach to analyzing the effectiveness of advertising. With the help of deep learning tools, in this project we identified the main objects present in the advertising video (clip); places where the action occurred; emotional component, age and sex of actors. Then we determined the combination of these elements in the frames and in the chronological sequence of the video, then this data were related to the category of the advertised object and the actual CTR. The rating of the most effective combinations of elements was compiled both in the frame and in the chronology of the video sequence.
Also, we conducted an experiment showing that based on the information extracted from the video, we can predict the ctr of the video using machine learning.
Second part of the study was to explore the possibility of application of our approach to the analysis of the popularity of a video posted to social networks, to predict the ratings on TV broadcasts and movies.
Many creative agencies and advertisers analyze the effectiveness of advertising materials, based on the results of past advertising campaigns. The analysis is rather laborious and occurs in a manual mode. Our approach allows automating the collection and allocation of the main patterns from the videos with the help of in- depth training and linking them to the indicators of the placement of the advertising campaign. The availability of large data on the effectiveness of certain elements of 
the video will allow advertising agencies to develop a more effective creative strategy and reduce the costs of the advertiser. The uniqueness of the research is also based on the availability of the largest database of video advertising in Russia. The study analyzed the advertising materials of the leading brands located on the Russian market.

Date: October 11, 2018