Ad Auctions for LLMs via Retrieval Augmented Generation

Authors: MohammadTaghi Hajiaghayi, Sébastien Lahaie, Keivan Rezaei, Suho Shin

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental An empirical evaluation validates the feasibility and effectiveness of our approach over several ad auction scenarios, and exhibits inherent tradeoffs in metrics as we allow the LLM more flexibility to allocate ads.
Researcher Affiliation Collaboration Mohammad Taghi Hajiaghayi University of Maryland hajiagha@umd.edu Sébastien Lahaie Google Research slahaie@google.com Keivan Rezaei University of Maryland krezaei@umd.edu Suho Shin University of Maryland suhoshin@umd.edu
Pseudocode Yes Figure 2: Single allocation segment auction; Figure 3: Multi-allocation segment auction; Figure 5: Naive II mechanism
Open Source Code Yes Github repo will be available.
Open Datasets No The paper mentions using "publicly-available LLM APIs" (e.g., GPT4) and a "model from the sentence-transformers library" for experiments, and describes generated ad texts, but it does not provide concrete access information (link, DOI, citation) for a specific public dataset used for training or as input data for the main experiments.
Dataset Splits No The paper describes running experiments over 500 trials with defined scenarios, but it does not specify traditional training/validation/test dataset splits or cross-validation details for its experimental setup.
Hardware Specification No The paper states: "All our experiments run with standard computation resource, without GPU." This is too general and does not provide specific hardware models, processors, or detailed specifications.
Software Dependencies No The paper mentions using "publicly-available API access of GPT4" and "a model from the sentence-transformers library" but does not specify version numbers for these or any other software components.
Experiment Setup Yes Setup. We consider segment auctions where each segment is a single sentence, and the entire document consists of three segments. There are five types of auctions: (1) segment auction with replacement... The query is fixed to: Can you suggest some books similar to To Kill a Mockingbird ?... We use publicly-available API access of GPT4 as our LLM in our experiments. In this Section, we first provide the set of prompts used to generate output in our experiments.