SearchPaper
The SearchPaper takes a user question and uses LLM to convert it to search queries for the corresponding scholar search engine. Then it will conduct the search and return a paper collection Node with the top-ranking papers:
Interacting with the results
Display search queries
It’s often important to rewrite the original question in specific ways as search queries to get the best retrieval results. Here we can see what are the search queries used to retrieve the papers. In this case, the result uses two generated search queries to retrieve the papers, and it displays the merged result.
Sorting the papers
We can sort the papers in this node with its Relevance Score (see the next section), Citation Count or Publication Year.
Labeling paper Relevance
It’s an important way to provide feedback to the search agent by labeling the relevance scores of each returned papers. Simplify hover over your mouse to Relevance 1 for each paper node, and it will pop up the 1 - 5 scale for you to select.
The rubrics are as follows:
- 1 = not relevant at all
No connection to the topic or question - 2 = a bit relevant
Slight relevance; mentions topic-related concepts but lacks useful content. - 3 = somewhat relevant
Indirectly helpful; provides background or related information - 4 = relatively relevant
Substantially aids in answering the question, though not fully comprehensive; or it might come from less recognized publishers - 5 = highly relevant
Directly answers the question with comprehensive information from a credible source
Follow-up Actions
Show more papers
Overall you find the thread of search is useful, and you want to continue with this line of search. You can click the Show more papers button, which uses the same search query to retrieve more papers.
AI-suggested search
Sometimes you find the set of search queries less helpful and you want to start a new thread of search. But comparing from starting completely from scratch, you can also attempt to prompt the models to come up with other queries given the current search results and your feedback. It uses the general search agent to generate multiple candidates for searching next.
User Suggestion
In this case, you want to provide your own search suggestions to the agent to plan for the new search. You can click the User Suggestion button, and it will pop up a dialog for you to enter your own search queries. You can similarly hit Shift + Enter to execute the search.
