Impact of Co-occurrence on Factual Knowledge of Large Language Models

TL;DR

This paper investigates how co-occurrence bias in training data causes large language models to make factual errors, especially with rare facts, and proposes debiased finetuning to mitigate this issue.

Contribution

It identifies co-occurrence bias as a key factor in factual inaccuracies of LLMs and demonstrates that debiased finetuning can reduce this bias but has limitations with unseen facts.

Findings

LLMs prefer frequent co-occurrences over correct answers.

Co-occurrence bias persists despite larger models or finetuning.

Debiased finetuning improves recall of rare facts seen during training.

Abstract

Large language models (LLMs) often make factually incorrect responses despite their success in various applications. In this paper, we hypothesize that relying heavily on simple co-occurrence statistics of the pre-training corpora is one of the main factors that cause factual errors. Our results reveal that LLMs are vulnerable to the co-occurrence bias, defined as preferring frequently co-occurred words over the correct answer. Consequently, LLMs struggle to recall facts whose subject and object rarely co-occur in the pre-training dataset although they are seen during finetuning. We show that co-occurrence bias remains despite scaling up model sizes or finetuning. Therefore, we suggest finetuning on a debiased dataset to mitigate the bias by filtering out biased samples whose subject-object co-occurrence count is high. Although debiased finetuning allows LLMs to memorize rare facts in the training set, it is not effective in recalling rare facts unseen during finetuning. Further research in mitigation will help build reliable language models by preventing potential errors. The code is available at \url{https://github.com/CheongWoong/impact_of_cooccurrence}.