The NeurIPS100 project is a newly launched platform-based intelligent mining service for top conference talents and top conference papers by AMiner. Its purpose is to provide in-depth insights into 100 authors and speakers (talents) of each top conference and analyze the relationship between authors. The relationship between the research factions, the growth path of the author, the forecast of future growth context, the job-hopping index, etc.; in addition, we will also conduct an in-depth interpretation of the 100 high-impact papers of the conference.
NeurIPS 2019 opened a few days ago. As one of the top conferences in the field of machine learning, every year there is a crowd of big cows.
And this year's number of participants reached 13,000. First, let's get a feel for the live picture, which is really bursting with popularity.
With so many participants and so many papers, how to efficiently obtain effective information from the conference has become a big problem.
• Among the more than 10,000 participants, how do you find your favorite mentor or collaborator?
• How to listen to valuable reports in hundreds of sessions?
• Among thousands of papers, how can I find the results I am interested in?
• Among the thousands of authors, who are the experts and rising stars in the field?
• What is their relationship in hundreds of technical fields?
• I have published/saw a good paper, how can I share it with interested people?
Aminer will help you solve the above problems!
AMiner has recently launched three major system tools: conference assistant conf-plus, KnoweldgeAtlas, and traceability tree , which will make your academic conference journey full of rewards.
The following scholars will introduce their functions and usage methods in detail.
No. 1 Conference Assistant conf-plus
（https://www.aminer.cn/conf/nips2019）
Conference assistant conf-plus, in addition to providing retrieval and reading of NeurIPS 2019 conference agenda, papers and other information, it can also view the latest news such as conference news reports, paper interpretation, and most importantly, it also provides detailed data analysis of selected papers.
1.Schedule : The detailed agenda of the meeting is introduced, including the topic, time, and location of each report. You can also click to view the corresponding papers and authors of the report;
2.Papers : Not only classified NeurIPS 2019 selected papers based on keywords, but also supports retrieval of paper titles, authors, and keywords;
Click on the author's name and a search box will pop up, showing all scholars with the same name retrieved by the system. You can vote to select the correct author of this paper.
When more than 10 people choose the scholar, the system will determine that the scholar is the author of the paper and tag him; at the same time, the author can also link to Xiaomai Scout to predict the scholar's probability of winning;
In addition, the award-winning papers of this year are also specially marked;
3.Like : Put a little like icon in front of each paper, you can like the paper you like, and you will collect your favorite papers in the Like column;
4. Statistics : Mainly show the data analysis of the authors and institutions of the selected papers in NeurIPS 2019;
5. Report : Contains relevant reports and paper interpretations of the NeurIPS2019 conference.
What everyone sees is the conf-plus1.0 version, and we will integrate more functions in the future to provide you with the most valuable services. Welcome to use and make suggestions.
No.2 KnoweldgeAtla
（http://knowledgeatlas.aminer.cn/）
KnoweldgeAtla is an intelligent knowledge system based on interaction, sharing and display.
Through the analysis of NeurIPS2019 papers and author data, we have generated a knowledge game that can better help users find papers and authors they are interested in.
Specifically, users can:
Acquire knowledge: check concept definitions and relationships, check well-known papers and authors, and give your own comments;
Check knowledge: question the concept that you think is incorrect;
Add knowledge: occupy new conceptual territory and nominate corresponding papers and experts;
Competitive knowledge : The leaderboard will show the user's score, which is more knowledgeable than one!
In addition to the ongoing NeurIPS, we will continue to launch the knowledge sharing of AAAI2020, WWW2020, ICLR2020 and other conferences in the later period.
No. 3 Traceability Tree (MRT)
Simply put, the tracing tree is a tool that can be used to help scholars study the evolution of a paper.
For a paper, the tracing tree searches its citations and automatically classifies them. Users can modify the generated traceability tree, and these records will be fed back to AMiner's algorithm later. At the same time, users can also download and share the traceability tree.
How to use traceability tree (MRT)?
By searching for the paper in AMiner, entering the paper page and clicking "Generate MRT" on the right, the user can initiate a request to generate this paper. Due to our limited resources, users currently need to share their generation request with at least two friends and invite them to jointly complete the creation of this tree. After that, our algorithm will take some time to complete the calculation. At the same time, it should be noted that the current algorithm cannot handle some papers that lack relevant information.
What algorithm is used?
In order to explore the evolution of a paper, we analyze the citations of the paper. Among them, the direct citation of the paper is an important clue, and its indirect citation can also reflect its historical development to a certain extent. Therefore, we selected the more influential part of these citations and used a series of related algorithms for analysis. .
Specifically, the algorithm extracts the semantic and structural features of these citations. The semantic features contain some information about the content of a paper, and the structural features introduce the topology of the citation network between different papers. Potential characteristics. We believe that semantically similar papers often share some topics, and the citation relationship reflects their developmental connection. After combining these features, we use an unsupervised learning clustering algorithm to divide these papers into several categories, and each category is displayed according to its time axis.
On this basis, we further perform semantic feature extraction for each category of papers to generate labels for that category. In the later period, we will further improve the algorithm, adding user feedback as semi-supervised learning information to the system to improve the clustering effect.
Everyone is welcome to use the above tools. If you have any questions during use, please leave a message to us.
(This article thanks Liu Xiao and Yin Da for providing reference materials, and thanks to zhengyiyu for the picture)
Past review: