Abstract
Evaluating advertising inventory value can be difficult, as methods that assess individual signals in isolation may not fully capture the quality of web traffic. This technology provides a method for generating an unified metric to assess the overall value of ad inventory. The system aggregates multiple behavioral scores, such as user engagement, ad load, and content quality, along with other signals like spam fraction. A machine learning model is trained to combine these inputs into a single, non-binary “combined behavioral score” for each website or application. The model is trained using a dataset of manually evaluated websites, with labels derived from quality metrics like deep conversion rates and lead quality metrics (LQM). This combined score provides an assessment of inventory value that directly correlates with advertiser outcomes, such as conversions. The score is computed on a daily basis, allowing the system to react to changes in publisher inventory. Keywords: Advertising inventory valuation, Behavioral score, Machine learning, Traffic quality, Ad fraud detection, User engagement, Conversion rate, Ad load, Spam fraction
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Sheng, Shuyang; He, Dongjing; Zippel, Richard; Tan, Xing; Dhuper, Sahil; and Fried, Zachary Loebel, "Machine Learning-Based System for Combining Behavioral Scores to Assess Advertising Inventory Value", Technical Disclosure Commons, (May 20, 2026)
https://www.tdcommons.org/dpubs_series/10186