Context Variance Evaluation of Pretrained Language Models for Prompt-based Biomedical Knowledge Probing.

Yao Z, Cao Y, Yang Z, Yu H. Context Variance Evaluation of Pretrained Language Models for Prompt-based Biomedical Knowledge Probing. AMIA Joint Summits on Translational Science proceedings. AMIA Joint Summits on Translational Science. 2023 Jun 16; 2023:592-601.

Dimensions for VA is a web-based tool available to VA staff that enables detailed searches of published research and research projects.

If you have VA-Intranet access, click here for more information vaww.hsrd.research.va.gov/dimensions/

VA staff not currently on the VA network can access Dimensions by registering for an account using their VA email address.    Search Dimensions for VA for this citation
* Don't have VA-internal network access or a VA email address? Try searching the free-to-the-public version of Dimensions

Search for Abstract from PubMed

---

Abstract:

Pretrained language models (PLMs) have motivated research on what kinds of knowledge these models learn. Fill-in-the-blanks problem (e.g., cloze tests) is a natural approach for gauging such knowledge. BioLAMA generates prompts for biomedical factual knowledge triples and uses the Top-k accuracy metric to evaluate different PLMs'' knowledge. However, existing research has shown that such prompt-based knowledge probing methods can only probe a lower bound of knowledge. Many factors like prompt-based probing biases make the LAMA benchmark unreliable and unstable. This problem is more prominent in BioLAMA. The severe long-tailed distribution in vocabulary and large-N-M relation make the performance gap between LAMA and BioLAMA remain notable. To address these, we introduced context variance into the prompt generation and proposed a new rank-change-based evaluation metric. Different from the previous known-unknown evaluation criteria, we proposed the concept of "Misunderstand" in LAMA for the first time. Through experiments on 12 PLMs, we showed that our context variance prompts and Understand-Confuse-Misunderstand (UCM) metric make BioLAMA more friendly to large-N-M relations and rare relations. We also conducted a set of control experiments to disentangle "understand" from just "read and copy".