CHLA 2023 Conference Contributed Papers/Congrès de l'ABSC 2023: Communications Libres
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Introduction:The use of controlled vocabulary to identify relevant articles is a central element of bibliographic database instruction in health sciences.Students learning to search MEDLINE are taught that MeSH yields precise results, and that MeSH indexing increases an article's findability, reliably describing an article's contents.Indexing for MEDLINE was done completely by human indexers until 2011.Since April 2022, all articles are assigned MeSH via automated indexing (AI).Per the NLM, MeSH assigned by AI are determined based on terms in title, abstract, and terms and indexing of 'related records', with human review and curation of results "as appropriate".As MEDLINE instruction typically starts with teaching learners to identify key elements or concepts in their research question and find appropriate MeSH for them, we sought to explore the following: how well does AI identify key concepts of an article?Are concepts missed more or less when compared to human indexers?Drawing on the PICO framework, are missing concepts more often any particular PICO element?Methods: We reviewed samples of automated and human-indexed records from shortly before April 2022, and some entirelyautomated from later, to determine whether their main concepts were adequately represented with MeSH.Working in pairs, our team used a web form to assign key concepts (based on the PICO framework) that, per our experience, would be used to find it and similar articles based on title and abstract.Assigned MeSH were then displayed and analyzed to determine whether they adequately represented the key concepts of each record.Results & Conclusion: As the study is ongoing, results are forthcoming.Potential impacts of Automated Indexing on library instruction and basic searching will be discussed. CP2. Can GPT-3 tools accurately find and analyze articles for systematic reviews? A (very) preliminary assessment Gary Atwood University of VermontIntroduction: GPT-3 is a large language model that uses artificial intelligence to generate textual responses to prompts and questions.GPT-3 technology has been used to create several interesting tools including the widely reported chatbot ChatGPT-3, which was released in November 2022.Inspired by the initial success of GPT-3, several organizations have started to build tools designed to assist with tasks associated with the systematic review research process.This project will analyze how successful these tools are in completing two specific tasks: searching for research articles and analyzing individual articles.Methods/Description: This project consists of two parts.In part one, the research question from a previously published systematic review will be used to conduct a search in two GPT-3 based tools for relevant research articles.In part two, each GPT-3 tool will be used to analyze a single research article to determine if it is relevant to the research question Results/outcomes: For part one, the results will be based on how effective and efficient each tool is at finding relevant research articles.Results will be compared to the set of articles included in the original review as a measure of success.For part two, each tool will be used to pull evaluative information from the sample article.This information will be compared to a manual assessment completed by the author.Discussion: This project will provide researchers with guidance on how to integrate GPT-3 based tools into their systematic review workflow.It will include a brief discussion of strengths and weaknesses and how they can impact potential results. CP3. Preliminary results of a longitudinal study of health information-seeking behaviour preand post-COVID
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.014 | 0.049 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.002 | 0.000 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.001 | 0.004 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it