NLM needs your input. We are experimenting with a new PubMed search algorithm, as well as a modern, mobile-first user interface, and want to know what you think. You can try out these experimental elements at PubMed Labs, a website we created for the very purpose of giving potential new PubMed features a test drive and gathering user opinions.
Please note that PubMed Labs includes only a limited set of features at this time and not the full set of PubMed tools. The absence of a feature or tool on PubMed Labs does not mean we plan to eliminate it from PubMed; it simply means we are not testing it now!
The key elements we are testing are:
- A new search algorithm for ranking (ordering) the best matches to your query
Based on analysis of data obtained from anonymous PubMed search logs, we have developed a new algorithm that we believe does a much better job of sorting search results by their relevance, or “best match,” to your query. This new algorithm incorporates machine learning to re-rank the top articles returned.
We were so excited by results with this algorithm that we already implemented it in PubMed, but it is still experimental and we would very much appreciate hearing what you think. Part of our test in PubMed Labs is having best match be the default sort, instead of PubMed’s default of sorting by most recent articles. If you find that you prefer to sort by the most recent articles instead, it takes only a simple click of a button to do so.
Interested in specifics about the new algorithm? You can read more in this NLM Technical Bulletin.
- Mobile-first, responsive design with modern user interface
PubMed Labs is designed to make searching and reading articles fast and easy, whether you are using a phone, tablet, laptop or desktop.
- Snippets from PubMed abstracts & highlighted search terms/synonyms
The search results page in PubMed Labs includes highlights (“snippets”) from the article abstract, when available, that are identified based on their relevance to the user query. Search terms and their synonyms are highlighted in both the title and the snippet.
We hope you take some time to try out the site and let us know what you think about these features. You can comment via this blog post as well as leave feedback via the PubMed Labs site.
I’m sorry that this comment is not on point but we want to capture someone’s attention at NCBI. These enhancements are great BUT the fact that you do not provide adjacency searching in PubMed at this point in the 21st century is bordering on negligence. My staff spends hours a week looking through PubMed citations seeing if an author is affiliated with a certain institution when an adjacency search function would automate this process. It’s a feature that Web of Science and Scopus off so we think the agency that has helped map the human genome could muster up enough tech help to provide adjacency searching in PubMed. Thank you, Heidi
Thank you for your comment. Please expect an email from the PubMed team as we are very interested in helping with your author affiliation searches.
I have tried more complicated strategies that we often deal in systematic reviews process and it´s work, but the result was less than the search in the original platform, I don´t know why. However my concern is to know more about the order of relevance in this machine. When I use a search strategy which combine MeSH terms with keywords in TIAB, or TW, and so on, which come first, for the relevance algorithm? Is MeSH more important that words in title, for example?
Thank you for your interest in the relevance order results. Please see the PubMed Help for an explanation of the best match algorithm: https://go.usa.gov/xn4q3
This is a nice demo and it looks like a lot of thoughtful work went into it. However, without allowing us to create collections, select individual articles, or access Loansome Doc, there’s a real risk that you aren’t going to get the level of beta testing you need. Speaking for myself, if I have no access to the features above, then testing this experimental interface is taking time out of my day to do something unproductive and not for my benefit. If you can link those features through into the demo, I would be happy to put the experimental search through its paces.
Thank you for your feedback. We’ll keep adding functionality and testing new functionality, so please check back for new and improved features.
I liked the fresh, modern look of the PubMed Labs interface, but in its present state it is lacking the features I rely upon and teach to students, such as the MeSH browser and article type filters, so it’s hard to say how usable it will be. I am curious as to whether search behavior, operators, and so on will change.
Thank you for your feedback. We’ll keep adding functionality and testing new functionality, so please check back for new and improved features.
Why do you need to change something that works so well? I don’t understand!
We’re glad PubMed works well for you. This is an experiment to see how we can make it even better.
Article 22357718 l am A Lamictal aseptic meningitis patients and need information. I’ve been off of it for ~18 months and I still live in hell
The best preparation for good work tomorrow is to do good work today and that what you really did.
we appreciate your hard work .
sci.najah.edu
I am late to the game but am finally looking at the new Best Match feature more closely: The new developments in PubMed using machine learning are impressive! I understand that the number of results when sorting by Most Recent or Best Match may be different, but what I’d like to clarify is whether the new Best Match algorithm may include records that are completely independent of the search query as entered by the user, i.e., records that are not in the Most Recent set at all?