All aggregators and database companies used by libraries are gradually integrating AI functionalities. They often do this by offering beta testers, usually significant clients, the opportunity to help finalize these new features. Others, not incompatibly, gradually make innovative tools available. This article looks specifically at Scopus AI, which came out of beta on 28 June 2024.Characteristics and content of Scopus AI
Scopus AI relies exclusively on the content of Scopus, specifically the abstracts (not the full texts of the documents) on which it was trained. For now, the documents considered do not date back further than 2013. A subscription to the Scopus platform is required to use Scopus AI.
Scopus provides access to over 91 million documents published since 1788, with references considered only since 1970. The referenced documents are of various types, mainly articles, conference proceedings, and book chapters.
Searching Scopus AI
Asking a Question Access to Scopus AI is through a tab on the Scopus home screen. You are then invited to ask a question in the search box. Scopus AI uses a combination of technologies, some from Elsevier and others not. It specifically uses a combination of its own LLM (Large Language Model) with others, including the GPT model developed by OpenAI. However, Elsevier specifies that this usage is strictly private—there is no information exchange that could be used to train the model.
Multilingualism Questions are asked in English since the abstracts in Scopus are in English, but there are projects to develop multilingualism. This already works, at least partially, we discovered. When we asked a question in French, it was perfectly understood, and the response was in English.
Structure of the response
The response consists of several elements. There is a brief summary followed by several paragraphs, each representing a thematic area of the response, with mentions of non-abstracted references on the right side of the screen. Finally, there is a quick conclusion indicating, where applicable, which questions did not receive a satisfactory answer. At this stage, up to ten references can be obtained.
Fundamental Documents: Below the references are the documents deemed fundamental. These have been cited by at least two of the references used for the presentation (summary plus paragraphs). They are often cited hundreds of times and can be quite old, sometimes predating 1973. Occasionally, the system may not propose any.
Extended Summary: There is an option to request an "extended summary" that offers additional analytical elements associated with references. At any time, references can be accessed with a single click in the classic search interface, allowing a seamless transition to this interface.
Concept Map: At the end of the first and second stages, Scopus AI displays a concept map, with the second potentially being more detailed than the first.
Experts: Another result often on offer is a list of up to three experts on the topic, along with their citation counts—we have seen more than 13,000—and a summarized presentation of their expertise. These experts are among the 19.6 million authors of publications referenced in Scopus. Sometimes the presented expertise may not be perfectly aligned with the question asked, however. Clicking on "preview profile" provides information on the recent interests of the expert and at least one reference related to the initial question. Clicking on "view full profile" leads to the full profile in the classic interface.
Go Deeper Section: At the bottom of the results is the "Go deeper" section, which provides three formulations of questions on specific aspects of the initial query. Curiously, these three questions appear even if the initial question did not find an answer, indicating that they are constructed from the initial query rather than its responses.
Reformulating the Query: It is worth noting that the desired results are not always found by the first formulation of the query, and it is common to have to restart the search with a different initial formulation, a classic approach with AI tools.
Target Audience: The exploitation of these results is intended to meet the needs of young academics or researchers working on various subjects. In other words, it is an excellent tool to quickly get a preliminary understanding of a topic and grasp its various aspects. However, Elsevier cautions that complementary research is recommended and that "Scopus AI is not intended to provide legal, financial, or medical advice."
Going Further: It should be noted that many cited references are not necessarily recent, which is inconvenient if you seek a "contemporary" answer to a question. Moreover, it is not clear how to use this tool for monitoring purposes.
More sophisticated uses
What has been described should be seen as a first step primarily aimed at non-experts of a subject to provide them with a structured and documented overview of the topic relatively quickly. This is clearly the core target as defined by Elsevier.
We aimed to go further by exploring how this new interface allows for addressing specific questions and setting up scientific/technological monitoring using Scopus's rich content.
The approaches proposed below may seem somewhat complex, but by combining a series of tools available in Scopus AI and the traditional platform, high-quality answers can be quickly obtained for both specific questions and monitoring setups.
Using Fundamental Documents: Start with the fundamental documents if they fit the initial issue, prioritizing the most cited ones, especially those cited hundreds of times or more recently.
Accessing the Complete Reference: Clicking on the article title displays the reference with the summary. A click on "open document detail" takes you to the complete reference in the classic interface.
Viewing Documents Citing the Reference: You can then click on "view 509 (in an example) citing documents." This provides a "classic" list of results with the bibliographic reference and analysis of these documents on the left side of the screen.
Processing the Results: There are several ways to process results. For monitoring purposes, if a significant number of documents are relevant, return to the previous screen and establish monitoring by clicking "set citation alert." This can be done from several fundamental documents or relevant documents in the results list. For a one-time search, if the number of documents is too high, add a criterion in "search in the results." The analyses of document sets in the left part of the screen can also be used, for example, by selecting keywords to retain or eliminate.
Choosing an Expert: Another approach involves using the expert list. Choose the most relevant expert, click on "preview profile," then "view full profile." This leads to the classic interface where the expert's profile is displayed, particularly their list of publications in "search result format." Select the relevant publications, then click on "cited by" to view the documents citing those selected. As above, there is the option to search within results and select and combine parameters displayed on the left side of the screen.
Setting Up Monitoring: For monitoring, follow the expert's publications and citations to their works.
Our opinion
As seen, in addition to using Scopus AI to quickly gain an overview of a subject, more sophisticated uses are possible, yielding very interesting results.