Local-Search Based Heuristics for Advertisement Scheduling

Mauro R. C. da Silva,Rafael C. S. Schouery

DOI: https://doi.org/10.48550/arXiv.2006.13432

2022-09-17

Abstract:In the MAXSPACE problem, given a set of ads A, one wants to place a subset A' of A into K slots B_1, ..., B_K of size L. Each ad A_i in A has size s_i and frequency w_i. A schedule is feasible if the total size of ads in any slot is at most L, and each ad A_i in A' appears in exactly w_i slots. The goal is to find a feasible schedule that maximizes the space occupied in all slots. We introduce MAXSPACE-RDWV, a MAXSPACE generalization with release dates, deadlines, variable frequency, and generalized profit. In MAXSPACE-RDWV each ad A_i has a release date r_i >= 1, a deadline d_i >= r_i, a profit v_i that may not be related with s_i and lower and upper bounds w^min_i and w^max_i for frequency. In this problem, an ad may only appear in a slot B_j with r_i <= j <= d_i, and the goal is to find a feasible schedule that maximizes the sum of values of scheduled ads. This paper presents some algorithms based on meta-heuristics GRASP, VNS, Local Search, and Tabu Search for MAXSPACE and MAXSPACE-RDWV. We compare our proposed algorithms with Hybrid-GA proposed by Kumar et al. (2006). We also create a version of Hybrid-GA for MAXSPACE-RDWV and compare it with our meta-heuristics. Some meta-heuristics, such as VNS and GRASP+VNS, have better results than Hybrid-GA for both problems. In our heuristics, we apply a technique that alternates between maximizing and minimizing the fullness of slots to obtain better solutions. We also applied a data structure called BIT to the neighborhood computation in MAXSPACE-RDWV and showed that this enabled ours algorithms to run more iterations.

Distributed, Parallel, and Cluster Computing

What problem does this paper attempt to address?

The problems that this paper attempts to solve are the maximum space occupation problem (MAXSPACE) in advertisement scheduling and its extended version (MAXSPACE - RDWV). Specifically:

1. **MAXSPACE Problem**:
- Given a set of advertisements \(A\), the goal is to select a subset \(A'\subseteq A\) from \(A\) and place these advertisements into \(K\) slots \(B_1, B_2,\ldots, B_K\), with each slot having a size of \(L\).
- Each advertisement \(A_i\in A\) has a size \(s_i\) and a frequency \(w_i\).
- A feasible scheduling scheme requires that the total size of advertisements in each slot does not exceed \(L\), and each advertisement \(A_i\in A'\) must appear in exactly \(w_i\) slots.
- The goal is to find a feasible scheduling scheme so as to maximize the space occupied in all slots, that is, to maximize \(\sum_{A_i\in A'} s_i w_i\).

2. **MAXSPACE - RDWV Problem**:
- MAXSPACE - RDWV is an extended version of MAXSPACE, introducing release dates, deadlines, variable frequency, and generalized profit.
- Each advertisement \(A_i\), in addition to having a size \(s_i\) and a frequency \(w_i\), also has a release date \(r_i\geq1\), a deadline \(d_i\geq r_i\), a profit \(v_i\), and upper and lower limits of frequency \(w_{\text{min}}^i\) and \(w_{\text{max}}^i\).
- Advertisement \(A_i\) can only be arranged in slot \(B_j\) that satisfies \(r_i\leq j\leq d_i\).
- The goal is to find a feasible scheduling scheme so as to maximize the total profit of scheduled advertisements, that is, to maximize \(\sum_{A_i\in A'} v_i w_i\).

To solve these problems, the paper proposes meta - heuristic - based solutions, including Greedy Randomized Adaptive Search Procedure (GRASP), Variable Neighborhood Search (VNS), and Tabu Search, and compares them with the existing Hybrid - Genetic Algorithm (Hybrid - GA). Through these methods, the paper aims to improve the efficiency and effectiveness of advertisement scheduling, especially in large - scale instances.