How to use artificial intelligence to help you efficiently find research topics?

Keep the spirit of exploration, learn to use the right tools, and solve problems efficiently.


For you, reading literature, especially literature that interests you, can be very enjoyable and exciting.

For example, Haihua, when he was a Ph.D. student, I went to UNT to study, and I was in a laboratory with him. He often reads the latest top conference papers, while admiring them. When he read something happy, he would shout it out, and excitedly pull me over to watch it together. I think he read out the mood of “A Song of Ice and Fire” through information retrieval.

But when it comes to reading the vast literature and trying to find your own research topic, it may not be so easy and enjoyable. The reason is clear to everyone – there are too many papers, and the speed of production is extremely fast. Even top-notch scholars dare not boast that they have read all the important literature in the field. Because just a few minutes after you speak, there may be (more than one) new paper published, which may include important new ideas and new discoveries.

This then constitutes an obvious contradiction:

The rest of my life has a career, but a dissertation has no limit. With the limitless, the game over.

However, without fully understanding the development of the field, how can you find and verify your own topic? You are excited to show others your new invention, a device that effectively reduces interface friction and improves transportation efficiency. Then others looked suspicious:

Isn’t this thing called a wheel?

For novices in scientific research, finding and selecting topics by studying papers is also responsible for the purpose of necessary scientific research training. Novices have no experience in choosing “which papers to read”, and the overall process may experience more setbacks and detours.

Some students are unwilling to spend useless effort on their own, and will simply push the problem to their tutors.

Teacher, please give me a reading list of papers.

If the supervisor is familiar with your research direction, this is a good thing to say. But if it happens that your research direction is a new field that your supervisor is trying to enter, then his help in this matter is limited. With so many new papers, he has not read all of them, so how can he tell which papers are more valuable?

In the end, it’s up to you to keep exploring in a fog. In the middle, you may miss a lot of important results, or even go in the wrong direction.

Fortunately, you can take full advantage of the tools.

The ancients sailed with a compass, but today you have navigation when you drive. Similarly, roaming in the world of scientific research, you can also use some more effective tools to reduce a lot of useless power consumption.


The artificial intelligence research aid I recommend to you is called Elicit. Link here .

I made a video for you on October 20, 2021 when I first discovered it.

In the video, I summarized several main functions that Elicit provided at that time, including:

  • Literature recommendation . After the selected topic is given, while recommending the literature, it also gives a claim;
  • Brainstorm . You give a topic, and after the AI ​​brainstorms, it gives you a list of possible topics;
  • Recommended by experts . Give the names of authors in one or two fields, and the AI ​​feeds back a list of authoritative experts in your field.

For demonstration of these functions, you can refer to the video content. I won’t go into details here.

What’s exciting is Elicit’s vision to provide researchers with direct and accurate query results . Elicit analyzes open literature libraries such as Semantic Scholar, and uses methods and technologies such as bibliometrics, information extraction, natural language understanding, and automatic summarization to master the in-depth features and associations of literature, so that you can conduct detailed queries.

Since Elicit uses the big language model GPT-3 as the driving engine, you don’t need to understand advanced retrieval methods, you can let Elicit understand what you are looking for through the dialogue form of natural language. This undoubtedly greatly reduces the training burden for researchers outside the field of graphics.

(Image credit: )

You might worry that literature databases like Semantic Scholar are not comprehensive enough. But a few days ago, at the “International Conference on Data Analysis and Information Service Development” held by our school, I asked Professor Kevin Boyack for advice. According to him, Semantic Scholar can be regarded as “a global model”, and the scale is sufficient.

Elicit has continued to evolve since my video last year. At the end of February this year, I also did a special live broadcast, using the question “How can a video be popular” as an example, to demonstrate some new features of Elicit.

You can click this link to view the live replay at that time.

In May of this year, I shared on the planet how to use Elicit for technology consulting .

After reading it, many readers gave enthusiastic feedback.

I found myself introducing and presenting at a rate that I couldn’t keep up with the Elicit evolution. As a (at least for now) free tool, the development team behind Elicit is really hard at work.

However, I also learned that many novice researchers do not understand the basic functions of this tool, and it is a pity.

Therefore I am writing this article for you. Let’s try to use a short space to focus on one question – how to use Elicit to help you find research topics.


Let’s open Elicit and enter a question first:

any recent interesting applications with GPT3?

After typing, click the query button, and Elicit quickly gives structured results.

If you skim through it, you will see that there is some “GPT” in the results here, which is not what we want. Because the emergence of GPT (Generative Pre-trained Transformer) in the field of artificial intelligence is a recent thing. And here are some results papers, published in 1997, on another GPT (General Particle Tracer).

What should I do if other topics are entered randomly in the search results?

For other, more tricky cases, we may need to use methods such as universal search to differentiate. But in our case, the simple and crude method is to filter by age .

According to the query, the time when GPT-3 appeared is in July 2020.

So, you can use the Filter on the right to set the publication time to after 2020.

Looking at the content of the results displayed on the left, there has been a significant change. The content filtered out this time is basically related to our problem.

But we do not know the importance of these articles. There are many indicators to measure the importance of an article, you can refer to my introduction in this article .

However, in order to simplify the problem, here we let Elicit do a simple sorting based on the number of references only.

After sorting, it becomes like this:

You can get an idea of ​​the research goals and a brief overview of these papers by looking at their titles and abstracts. Some of these articles can be directly opened as PDF files (blue links) to view specific research content.

It doesn’t matter that some search result items are not provided directly in PDF. As a last resort, you can also obtain the full text from the author directly through the Semantic Scholar platform. “As long as the mind doesn’t slip, the method is always more difficult than the difficulty.”

When reading PDFs, you might as well use the methods I mentioned to you before, and actively seek feedback from reading literature to improve your understanding of methods and paradigms in a certain field.

You can also add a series of elements (columns) to the list given by Elicit. For example, author, source journal, funding, and DOI.

However, if it only displays these metadata content, it is overkill for Elicit. You can let Elicit display the results of intelligent analysis, such as article categories, research methods, etc. For experimental articles, you can even extract information such as the number of samples, specific details of the samples, age, and region.

These data displays reflect the characteristics of Elicit’s “fine-grained” and “understood natural language”. It is no longer satisfied with providing “book cover kung fu”, but helps you chase it to the end, using the powerful functions of GPT-3 to complete information extraction and sorting.

You can try these functions one by one according to your own needs. If you don’t know the usage scenarios and usage methods of some functions, you can refer to a series of short instructional videos provided by Ought (Elicit Production Company). The video is not long. I have watched a few paragraphs in detail before and found it very rewarding.

Let’s talk about my favorite feature – let Elicit take advantage of the language understanding capabilities provided by GPT-3 and try to directly answer your research questions .


Remember what our problem was?

any recent interesting applications with GPT3?

This question, in fact, does not require you to go through the literature one by one to reach it. Because you can have Elicit add this:

As you’ll soon see, Elicit summed up our problem like this.

How about it? Great, right? Through natural language understanding and summarization, you ask questions, and Elicit directly answers them.

I need to warn you, though, not to have high expectations for Elicit’s direct answer . Some people make fun of the current level of artificial intelligence, which is “artificial mental retardation”. This is an ugly word, but it actually makes sense. After all, artificial intelligence research is still a long way from the goal of “strong artificial intelligence”. These automatically generated answers are likely to be imprecise and even contain errors. But in any case, it gives you a good foundation and starting point for further research.

Some students must be impatient at this time:

Teacher, what’s the use of saying so much? I just want to find a good topic to start with! How does it benefit me to find out so much relevant research and know what applications a certain technology has? ! I can’t do what others have done again.

First of all, other people’s research has reference value for you, and it is especially helpful to help you determine the necessity of many research directions.

Second, let’s not be so eager for quick success. Others spend months honestly doing a literature review, you expect Elicit to do it in seconds, is this realistic?

Hmm, hard to say.

I am in ” How to choose a research topic? The article told you that the “research limitations” of others may be a starting point for your continued research. Of course, you need to be careful not to fall into a hole dug by others.

The great thing about Elicit is that it can quickly list these limitations for you after “explore the group articles”. It is much more efficient than you rummaging and checking one by one.

We’ll add the “Study Limitations” column here.

Note that this column of results may not be comprehensive.

Some lines have no content at all. What’s going on here?

Which paper is not followed by a description of limitations? You can’t even draw this thing, how dare you call yourself smart? It seems that Elicit is really “artificial mental retardation”.

Don’t jump to conclusions yet. Look carefully at what the “limitations” are. It can be seen that these are not completely equivalent to the limitations described at the end of the paper, but are actually discussing the limitations of GPT-3 technology . This is ” highly relevant ” to your question. Those given answers, because of this high level of relevance, may be worth digging deeper .

In response to the limitations put forward by others, if you can give full play to your own strengths, combine your unique perspective, and provide valuable new answers, then victory will beckon you. But wait, remember I mentioned to you the pit of research limitations before? For example, what if these limitations are just foreshadowing by the author for his subsequent series of articles?

In fact, the answer is still Elicit. We can convert the limitations discovered by Elicit into new problems, and let Elicit help us find other people’s research results in the massive research library, so as to determine whether they are valuable real problems, or whether they have been preempted by others. In this way, you can avoid a lot of detours.

By analogy, try it.


Review the process we used Elicit in this example.

We start by asking a question that is not specific enough, or not confident enough. By filtering the results given by Elicit and adding elements, we can constantly iterate on our own problems, adjust the research angle, and finally find a “research niche” that can satisfy our own curiosity and be valuable enough to others.

Be careful not to set too high expectations for finding the right topic overnight. Because then you are more likely to experience frustration. If you quickly identified a high-value research question, congratulations, good luck! It doesn’t matter if there are twists and turns in the middle. These continuous attempts are also giving you the knowledge to accumulate research topics and research results in this field. Elicit has actually helped you quickly obtain a clear and complete research map.

Due to space limitations, other useful functions of Elicit are not covered in this article. Given that Elicit is frequently updated, by the time you read this article, new useful features may be available. Keep the spirit of exploration, learn to use the right tools, and solve problems efficiently.

Do you have any better methods and techniques for studying scientific literature? Is there a better tool to recommend? Welcome to leave a message, share your experience, we will exchange and discuss together.

Please charge if you find this article useful.

If this article might be helpful to your friends, please forward it to them.

Welcome to my column “Scientific Research Tool” to receive follow-up updates in a timely manner.


Further reading

Title image: Drawing by OpenAI DALLE 2. prompts: a scientist facing towards the gate of a giant maze

This article is reproduced from:
This site is for inclusion only, and the copyright belongs to the original author.

Leave a Comment

Your email address will not be published.