Streamlining literature review with agentic AI in the Web of Science

Powered by an advanced AI agent, the new literature review guide in the Web of Science Research Assistant helps researchers conduct multi-step, complex reviews with greater accuracy and speed.

When the Web of Science team began exploring ways that generative AI could improve academic research, we quickly identified its potential to help researchers complete complex tasks faster. In the Web of Science Research Assistant, prompts and simple guided walkthroughs help researchers identify new angles on research topics and next steps for their work. By implementing agentic AI-the next wave of AI technology- in the assistant, we can provide even deeper support for researchers carrying out multi-step, involved research tasks.

We are pleased to announce a redesigned literature review guide in the Web of Science Research Assistant: Literature Review 2.0. Powered by an advanced AI agent, the guide uses powerful reasoning and problem-solving skills to help researchers quickly refine their search strategies, define the scope of their tasks, and curate the outputs provided by the assistant.

The benefits of AI agents for academic research

The first wave of generative AI-powered chatbots and assistants leverage their language parsing and search indexing capabilities to gather results and recap key takeaways as quickly as possible. Researchers benefit from a more natural search and discovery experience, but this can come with a drawback: responses sometimes may not meet user expectations. Because these tools respond in a transactional manner without first clarifying or validating user needs, included topics might be too broad, and documents too old or too recent. It might take a researcher several attempts to get the information they seek.

In contrast, agentic AI gives large language models the power of reasoning. AI agents can plan and execute multi-step processes by interacting with users, data sources and tools. This benefits researchers with nuanced questions that require careful term selection and consideration of document characteristics for a relevant response. Guided by close collaboration with the user, the agent evaluates its assigned task, designs a solution and determines when it has enough information to complete the work. Agents share and validate strategies and methods with the user, bringing transparency to the process.

Creating a more interactive literature review process

For researchers conducting a literature review, conversational exchanges with an AI agent can lead to highly relevant papers and valuable insights faster. Our redesigned literature review guide enables users to work with an agent powered by responsible Academic AI and trusted Web of Science Core Collection data to identify knowledge gaps, locate research hotspots and formulate hypotheses. The agent converses with researchers to understand their intent and preferences, then determines the best approach for conducting a literature review based on their specific needs. This personalized support creates an interactive experience that more closely mimics working with a human assistant compared to what generative AI alone can offer.

Through a back-and-forth exchange, the literature review agent defines the research question, confirms the search strategy, and then delivers a report based on various analyses of the literature. At each point in the process, the user provides input on exactly what they are looking for. Working with the agent, users can:

• Optimize the query: The agent analyzes the research question and determines if it would benefit from being more specific, offering intelligent suggestions of subtopics to narrow the search, if needed.
• Customize the scope: Taking the refined research question, the agent will propose a search strategy, listing the themes and keywords, including synonyms and related terms. Transparency helps the user further refine the search by excluding or adding terms and specifying a date range.
• Define the output: After validating the search strategy, the agent will confirm the structure of the report it will generate. The user can specify any sections to exclude, which to expand or even how many bullet points they would like in a particular section.

In addition to customizing the literature review process to meet the unique needs of each Web of Science Research Assistant user, the agent also delivers new types of information about the research landscape. It streamlines the process of synthesizing the literature and distilling the most salient elements, unlocking greater research efficiency while keeping researchers and their valuable expertise central to the process.

Figure 1: Work with the agent to define your search strategy around what matters to you

Figure 2: Get the information you need structured how you want it, including common themes, research gaps and hypotheses

The future of agentic AI in the Web of Science

The Web of Science Research Assistant is designed to help researchers quickly carry out tasks that usually take hours or days to complete, and agentic AI presents numerous exciting opportunities for its future development. In addition to the new literature review guide, we plan to add additional capabilities leveraging its strengths over time, such as options for creating alerts, exporting to various formats and products and expanding discovery to more output types. We are committed to empowering researchers by improving their day-to-day research experience, and continue to develop new features in collaboration with the research community, engaging customers as we validate and test new features and ideas.

See a demo of the new agentic-AI powered literature review guide in action.