archAIve

• 

  Output for query "quantum physics"

Inspiration

We chose this problem to make navigating research literature less overwhelming and more accessible. This recommender system helps users efficiently find relevant papers, save time, and improve access to related research.

What it does

archAIve is an AI-powered research article recommender system to help researchers and students discover relevant papers based on their interests. The system takes a user-provided query as input to recommend the most similar papers from the ArXiv dataset with their author(s), category tag and click-able link to the full paper.

How we built it

Initially, we used a TF-IDF vectorizer with a KNN search to find similar articles based on cosine similarity. However, to improve performance, we increased the size of our dataset, switched to dimensionality reduction with SVD, and implemented FAISS for fast approximate nearest neighbor retrieval. This allows the system to efficiently recommend the top 5 papers with highest relevance.

Accomplishments that we're proud of

A satisfying result is that the cosine similarity score is consistently above 0.8. Also, queries with higher similarity scores produce near-perfect Precision@5, which indicates that the top 5 recommended papers are actually relevant to the query.

What's next for archAIve

A continuation of this project would be to extend it to include preprints for more scientific domains, by combining other preprint datasets (ex. bioRxiv). Another improvement would be to incorporate user feedback into the system by allowing users to rate how useful or accurate each recommendation is, “like” or “save” specific papers, and track what they tend to click on. This feedback could then be used to fine-tune the model and personalize the search results for each user based on their past preferences and interaction history.

Built With

• arxiv
• faiss
• html
• matplotlib
• meta
• numpy
• pandas
• python
• scikit-learn