Ranking with Multiple Objectives

Nikhil R. Devanur,Sivakanth Gopi

2024-10-16

Abstract:In search and advertisement ranking, it is often required to simultaneously maximize multiple objectives. For example, the objectives can correspond to multiple intents of a search query, or in the context of advertising, they can be relevance and revenue. It is important to efficiently find rankings which strike a good balance between such objectives. Motivated by such applications, we formulate a general class of problems where - each result gets a different score corresponding to each objective, - the results of a ranking are aggregated by taking, for each objective, a weighted sum of the scores in the order of the ranking, and - an arbitrary concave function of the aggregates is maximized. Combining the aggregates using a concave function will naturally lead to more balanced outcomes. We give an approximation algorithm in a bicriteria/resource augmentation setting: the algorithm with a slight advantage does as well as the optimum. In particular, if the aggregation step is just the sum of the top k results, then the algorithm outputs k + 1 results which do as well the as the optimal top k results. We show how this approach helps with balancing different objectives via simulations on synthetic data as well as on real data from LinkedIn.

Data Structures and Algorithms,Discrete Mathematics

What problem does this paper attempt to address?

### What problem does this paper attempt to solve?

This paper mainly focuses on the problem of how to optimize multiple objectives simultaneously in search and ad ranking. Specifically, the authors have studied how to effectively find a ranking method so that a better balance can be achieved among multiple objectives. For example, in a search engine, a user may have different intentions for the same query term (such as "jaguar" which can refer to an animal or a car company), so a ranking method needs to be found to satisfy these different intentions. In addition, in an advertising system, there are usually two main objectives: relevance and revenue. The trade - off between these two objectives is crucial for improving user experience and advertising effectiveness.

### Main contributions of the paper

1. **Problem modeling**:
- Each result has a different score for each objective.
- Ranking results are aggregated by weighted summation, and the weights can decrease according to the position.
- Use an arbitrary concave function to maximize the combination of these aggregated scores.

2. **Algorithm design**:
- Proposed an approximation algorithm that can work in a bi - criteria/resource - augmented setting. This algorithm is slightly better than the optimal solution, especially when only considering the top k results, the k + 1 results output by the algorithm are no worse than the optimal k results.
- Verified the effectiveness of this method on synthetic data and real - world LinkedIn data through simulation experiments, demonstrating its ability to balance between different objectives.

3. **Theoretical guarantee**:
- Proved that when using a concave function to combine multiple objectives, a more natural balanced result can be achieved.
- Provided a bi - criteria approximation algorithm with polynomial - time complexity, which is suitable for various application scenarios.

### Markdown representation of formulas

- Cumulative score formula:
\[
csw(a_i, \pi) = \sum_{j \in [n]} w_j a_{\pi(j)}
\]
\[
csw(b_i, \pi) = \sum_{j \in [n]} w_j b_{\pi(j)}
\]

- Maximized objective function:
\[
\max_\pi f(csw(a_i, \pi), csw(b_i, \pi))
\]

- Definition of weight vector:
\[
w_i = \frac{1}{\log_2(i + 1)}
\]

- Global cumulative score:
\[
gcsw(a, \pi) = \sum_{i = 1}^m csw_i(a_i, \pi_i)
\]
\[
gcsw(b, \pi) = \sum_{i = 1}^m csw_i(b_i, \pi_i)
\]

Through these formulas and methods, the paper proposes an effective and flexible way to deal with the multi - objective ranking problem, ensuring a good balance between different objectives.