-
-
Newsletter
Inspiration
The inspiration for our research paper recommendation system came from researchers struggling to keep up with the vast amount of information in their respective fields. We wanted to create a solution that could alleviate the time-consuming task of manually searching for relevant research papers and provide users with tailored recommendations based on their interests. This would help them learn about new advancements in their field without much effort. Try it out here: https://papyrai.shiporium.com/!
Additionally, we wanted a simple way to give people summaries of research papers without them needing to read too much of the paper, as we found that whenever we saw a research paper, even the abstract was often much longer than we wanted. That's why we wanted to add an LLM to automatically filter the most useless information for a quick read.
What it does
Our research paper recommendation system utilizes advanced LLM algorithms (including GPT-3.5) to analyze user preferences, fetching research papers from reputable sources. By considering various factors such as topic relevance and author credibility, our system generates personalized recommendations that match the user's interests provided when they sign up. These recommendations are then delivered to the user through an intuitive email, with an intuitive interface for viewing links. We have also implemented a system to track exactly what links the user clicks on, to track what they are interested in and adjust our recommendation algorithm based on that information.
How we built it
We built our research paper recommendation system using a combination of GPT-3.5 and prompt engineering, APIs of multiple research journals, and SMTP email sending. We used prompt engineering to summarize and categorize research papers based on their content and topics. We also developed robust data-standardizing algorithms to extract relevant information from various research paper databases. The backend was built using a scalable architecture that ensures efficient storage, retrieval, and recommendation generation.
Challenges we ran into
During the development process, we faced several challenges. One of the main hurdles was acquiring and processing a large volume of research paper data from different sources, which required significant effort in data cleaning and preprocessing. Additionally, fine-tuning the recommendation algorithm's prompts to balance accuracy and relevance was much more difficult than we previously thought, and required a lot of edge cases to be collapsed.
Accomplishments that we're proud of
We are proud to have created a research paper recommendation system that simplifies and enhances the research process for users. Our system tracks the actions of the user once they are sent an email, allowing the app to track exactly what they are interested in and provide more accurate and personalized recommendations over time.
What we learned
Throughout the development of this project, we gained valuable insights into prompt engineering, API usage, and recommendation systems. We learned how to effectively preprocess and analyze large datasets, implement complex algorithms, and optimize performance. We also developed a deeper understanding of user experience design and the importance of delivering personalized and relevant content.
What's next for Papyr AI
In the future, we envision expanding the capabilities of our research paper recommendation system by incorporating user feedback and preferences to enhance our AI model's recommendation accuracy. We also plan to integrate collaborative filtering techniques and a social media system, allowing users to discover papers based on the interests and recommendations of like-minded researchers. Additionally, we aim to integrate with academic platforms and provide a seamless experience for users to access and organize their recommended papers directly from within the system - including browser extensions, reading lists, and more!
We strongly believe that this program can soon evolve into something that can provide researchers and academics an effortless and seamless way to consume, browse, and explore research.
Log in or sign up for Devpost to join the conversation.