Performance of Native and Catalogue Advertising

The dataset contains large-scale, impression-level advertising data collected from performance-oriented e-commerce campaigns delivered on a major digital advertising platform. Each observation corresponds to a single advertisement impression, defined as one exposure of an advertisement to a user at a specific point in time within an algorithmically curated content feed. The dataset integrates advertising exposure logs with anonymized user interaction signals, creative-level metadata, inferred customer journey states, and downstream behavioral outcomes. The data include impressions from two distinct advertising formats: sponsored (native) advertisements, which embed promotional content within the surrounding informational environment, and catalogue-based advertisements, which dynamically present specific products based on structured product feeds and user behavioral history. For each impression, the dataset records the advertising format, timestamp, anonymized user identifier, inferred journey stage (early, mid, late), behavioral features capturing recency, interaction depth, and product engagement, as well as binary outcome variables indicating clicks and post-click conversions.
For native advertisements, additional semantic features are provided, including measures of semantic similarity between the advertisement and its contextual environment, content–topic overlap, and sentiment alignment. These features are derived from multimodal content representations of advertising creatives and surrounding content, enabling analysis of semantic congruence and contextual relevance. Customer journey stages are inferred probabilistically from pre-exposure behavioral patterns using time-decayed interaction features, reflecting evolving user decision states rather than predefined marketing funnel labels. The dataset is designed to support research on human–algorithm interaction, digital persuasion, and advertising effectiveness across user decision stages. It enables both descriptive and model-based analyses of how semantic and behavioral signals interact with user state to shape engagement and conversion behavior in algorithmically mediated environments.