I’ve always wanted a simple semantic search interface for an ACL event page (e.g. EMNLP 2024) to find relevant papers beyond searching by keywords or skimming the abstracts. This is especially useful for finding related work for a specific task from recent ACL conferences.
With ACL Event Search, you can do just that by simply providing the url of an event page. It is an open source tool which fetches all papers from the event page, creates an embedding index using sentence-transformers, and lets you perform semantic search. Moreover, it also clusters all the abstracts and titles to identify the variou tasks tackled by the accepted papers of an event. I experimented with the Gemini API for generating the cluster labels and it works quite well. As I use HDBSCAN for clustering, it is possible that a large number of papers can sometimes be grouped under the “Uncategorized” cluster (noise points). The tool simply lets you rerun the clustering on these large clusters to find specific subtopics.
Below are some screenshots of ACL Event Search in action:
Load all papers from an event page. Clustering is performed on the abstracts and titles in the background.
Semantic search using natural language queries.
Cluster labels generated using Gemini (here 1.5-pro) denote the various research tasks.
Example cluster with related papers.
Fetched papers for each event are cached along with their embeddings and the generated cluster labels. Reclustering is quite fast but potentially expensive if done multiple times. The tool is open source and can be found on GitHub.