# Research Paper Analysis Visualization

## Introduction

## Related works
A Visualization of Research Papers Based on the Topics and Citation Network [[IV 2015](https://www.semanticscholar.org/paper/A-Visualization-of-Research-Papers-Based-on-the-and-Nakazawa-Itoh/7b5956d72712563470bd4fd84324d6c935f36686)

RTVis: Research Trend Visualization Toolkit [[Demo]](
https://www.rtvis.design/) [[Arxiv 2023]](https://www.semanticscholar.org/paper/RTVis%3A-Research-Trend-Visualization-Toolkit-Shen-Lin/d1d5be861d7db45125f31dbef14dfa23126eb4d8)

[Elicit](https://elicit.com/): A LLM enhanced tool used to search papers for a specific problem

## Traditional Way to Look for Papers
When do we survey papers and how do we do it?

### 1. Intentions
Why and when do one need to look for papers? I concluded the common use cases into several categories:
(1) Having little knowlege about a certain topic, trying to obtain an abstract insight of the core concept.
(2) Already have a certain domain knowlege in the topic, trying broaden one's view to its related topics.
(3) Trying to obtain the current trend of a specific topic, and catch up with the newest findings.
(4) paper's similarity with our work
(5) compare advantage/disadvantage

Below are the common ways to accomodate the different needs above:
1. Read through the survey papers covering the certain topic.
2. Read through the prior/derivative works with similar topics but different perspectives.
3. Search through the prior/derivative works, find the trending of different keywords through time.

### 2. Traditional Searching Methods
Below is the traditional way for how one surf through the piles of papers to decide which ones are important to them.

(1) With some ideas and keywords in mind, search by specific keywords.
(2) Starting from specific papers, look for the related works. (prior/derivative/side-related)

1. By keyword
* Type in relevant words in Arxiv/Google semantics search prompt.
2. By author
* Find relevant works by the first/last author name.
3. By reference
* Going through the referenced papers by the current given papers.

### 3. Decision (Reading Priority)
One's time and interest are limited. It is impossible for us to read through all of the papers. It is essential for us to pick out the more important ones to read.

Below are the factors that may be used to decide whether a paper is more important for us:
1. By relevance (between the provided papers and keywords)
2. By citation
3. By publish date
4. By conference/author name

### 4. Where to Search
Arxiv
IEEE
Google Scholar
Semantic Scholar

-----
以下開始是我們該如何去實作，
有想法拜託打上去

## 構想

在沒有domain knowledge的前提下，要一次搜尋到位不太實際。
目標應為：讓使用者可以在使用工具搜尋的過程中得到insight，並經過反覆幾次搜索(更換關鍵字、屏蔽部分結果/關鍵字...)，以獲得最終想要的內容。

* 使用流程假想圖
--> ***初次使用，丟進一些keywords/paper***
--> 獲得結果，並得到一些insight
--> 篩掉一些不相關的結果，並更新使用的keywords

--> 獲得結果，並得到一些insight
--> 篩掉一些不相關的結果，並更新使用的keywords

--> 獲得結果，並得到一些insight
--> 篩掉一些不相關的結果，並更新使用的keywords

--> 獲得結果，並得到一些insight
--> ***初步認識題目，依據重要性(or 推薦路徑)開始閱讀論文***
--> ***對相關領域有更深入的理解***
--> 篩掉一些不相關的結果，並更新使用的keywords

--> 獲得結果，並得到一些insight
--> 篩掉一些不相關的結果，並更新使用的keywords
--> and so on ...

## 研究人員使用這個工具的目標：
1. 回答使用者的研究問題
2. 對該領域一無所知，希望能在使用這個工具的過程中，得到該領域相關的一些knowledge and insight
3. 讓使用者想更深入了解某個領域，可以以此工具獲得一個大略的paper重要性&閱讀順序
4. 讓使用者可以依照他看過的paper，以此工具比較相關的works及他們的優劣
5. [Leverage predictable pattern of research progress](https://youtu.be/P7AoHd2nC8g?si=-Lv0PZ6W9T179_nI&t=5160)
6. 某領域中新/舊的work

## 所需要的功能：
* By LR：
* Basic keyword-based search (by calling API)
* 使用者可以改變node/edge的color pallete (以highlight重點)
* 提供使用者一些意見
* 根據使用者搜尋的結果，建議下次搜尋使用的keyword
* 可以篩選/屏蔽 conference name、author name、keyword name
* 區分出不同類型的paper
* survey(review or tutorial)
* theory
* implementation-oriented
* etc ...
* [LLM-powered Intellegent Search](https://ambersearch.de/keyword-based-vs-federated-vs-intelligent-search/#:~:text=A%20keyword-based%20enterprise%20search%20works%20according%20to%20the,assess%20the%20relevance%20of%20the%20hits.%20...%20%E6%9B%B4%E5%A4%9A%E9%A0%85%E7%9B%AE)
* Cache住上次的紀錄
* word search, highlight nodes with the word included
* By

## 如何Demo
* 可以得到的insight
* 不同的paper類型
* 冷門/熱門的領域(keyword)
* 冷門/熱門的研究題目(可能更細的keyword?)
* 不同年代面對/提出的研究問題，後來如何解決及如何發展
* ...

* Use Cases
1. 使用者原本有個粗略的研究方向，使用工具後方向訂得更明確了
2. BBB

## 如何實作
* Library - React
* Framework - Next.js
* Deploy - Vercel
* Data - Semantic Scholar API
* Visualization
* Force Directed Graph
* [Nivo network](https://nivo.rocks/network/)
* [react-force-graph](https://github.com/vasturiano/react-force-graph?tab=readme-ov-file)
* Bar Charts
* ...
* Algorithm
* LDA
* ...

## 12/18 Discussion

* Citation Network
*
* Keyword Relevance Network

* 使用步驟
1. 使用者用關鍵字搜尋 paper 得到 top20 paper (A)
2. 抓 A cite/cited paper 各5篇 (B)
3. 建立 group1: {A+B}
4. 畫出 group1 的 citation network
5. 推薦使用者新關鍵字: LDA(group1)

5. 每次關鍵字搜尋會建立新的group
6. 使用者可以delete/modify/merge group

### 分工
Deadline 12/25(一)
* 王培碩
* Draw simple citation network
* highlight/(un)fold
* cite direction
* 陳盈圖
* API
* LDA
* 楊秉宇
* 會實作 space-filling algorithm
一張圖中 $G = \{N,L\}$ 其中 N 是的論文群的數量 用 $N=\{n,...,n_k\}$ ，$k$ 是論文群的數量
每個論文群有一個 $m$ 維度 0-1值，用 $ni = \{b1,...,bm\}$ 假設 $b_i = 1$表示 這個論文群是屬於第 $i$ 個 categories。

Deadline