Continuous calculation method of brand advertisement effectiveness optimization

[0075] Use the ad click log, ad display log, creative information, and user information in the video website for a day to integrate and clean the data, and then extract and format the features. A total of 35 features are extracted, of which context features include: time, city, device type, etc.; creative features include: material industry, sub-industry, category; user information features include: gender, age, advertising preference, channel preference; combination features include : Time and device type combination, gender and advertising industry combination, etc. The extracted features are discretized by One-hotencoding, so these 35 features are mapped to a 5369-dimensional feature space. Each sample will be represented as a 5369-dimensional vector and a record identifying itself as a positive or negative sample. Since 5369 dimensions is a large feature space, and the entire data matrix is ​​very sparse, only the feature dimensions with values ​​are stored. Such as image