Efficient and Practical Approximation Algorithms for Advertising in Content Feeds

TL;DR

This work tackles native advertising in content feeds under decaying user attention by formalizing two offline problems, $ extsc{StrmAds-R}$ and $ extsc{StrmAds}$, with a quitting probability $q$ and reward structure $f(M)= obreak \sum_{e=(i,j) obreak olinebreak  M} r_{e}igl(1-qigr)^{j+z(j)}$. It introduces fast, provably $2$-approximation greedy algorithms that emphasize bottom-slot decisions through backward processing, notably the optimal $ extsc{G-bwd}$ for StrmAds-R and the non-oblivious $ extsc{G-bpx}$ for StrmAds, along with a natural variant achieving the same guarantee. The paper also presents additional practical algorithms (Flow, FlowG, MWM, and a lazy $ extsc{G-glb}$) and an online set of heuristics, accompanied by the first comprehensive empirical evaluation on synthetic and real data demonstrating robust performance and scalable runtimes. Overall, the results show that backwards, bottom-slot–aware strategies yield strong practical performance with solid theoretical guarantees for native streaming advertising in content feeds.

Abstract

Content feeds provided by platforms such as X (formerly Twitter) and TikTok are consumed by users on a daily basis. In this paper, we revisit the native advertising problem in content feeds, initiated by Ieong et al. Given a sequence of organic items (e.g., videos or posts) relevant to a user's interests or to an information search, the goal is to place ads within the organic content so as to maximize a reward function (e.g., number of clicks), while accounting for two considerations: (1) an ad can only be inserted after a relevant content item; (2) the users' attention decays after consuming content or ads. These considerations provide a natural model for capturing both the advertisement effectiveness and the user experience. In this paper, we design fast and practical 2-approximation greedy algorithms for the associated optimization problem, improving over the best-known practical algorithm that only achieves an approximation factor of~4. Our algorithms exploit a counter-intuitive observation, namely, while top items are seemingly more important due to the decaying attention of the user, taking good care of the bottom items is key for obtaining improved approximation guarantees. We then provide the first comprehensive empirical evaluation on the problem, showing the strong empirical performance of our~methods.

Figures (6)

• Figure 1: An illustration of the expected reward being non-monotone with respect to the ad placement. Here $a,v$ and $s$ denote ads, videos, and slots respectively. In the first scenario (a) an ad $a_2$ with reward 2 is allocated to slot $s_2$ after video $v_2$, and a user sees the ad $a_2$ with probability $p$. In the second scenario (b) an additional ad $a_1$ with reward 1 is allocated to slot $s_1$ after video $v_1$. Due to decaying user attention, in (b), the user sees the ad $a\xspace_2$ with a probability $p'<p$. Thus, placing an additional ad may lead to a smaller expected reward.
• Figure 2: Representation of \ref{['exmaple:top']}, where a natural online greedy algorithm and maximum weighted matching ($\textsc{MWM}$) perform poorly ($a\xspace$ and $s$ represent ads and slots).
• Figure 3: Comparisons on synthesized bipartite graphs with different weighting schemes. Error bars indicate the standard deviation.
• Figure 4: Effects of parameters $n\xspace, m\xspace, q\xspace, k$.
• Figure 5: Comparisons on simulated native advertising using real data.

Theorems & Definitions (2)

• Example 1
• Example 2: Being myopic in top slots