Datasets:

Modalities:
Text
Formats:
parquet
Libraries:
Datasets
Dask
License:
File size: 878 Bytes
add98dd
 
 
3b36ff1
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
---
license: apache-2.0
---

# Subjects-for-curricular

---

Subjects-for-curricular is a clustered set of book titles and concepts from https://huggingface.co/datasets/benxh/opensyllabus-tagged-libgen and https://huggingface.co/datasets/benxh/us-library-of-congress-subjects.


These two datasets are combined, taking the 'text' field from the US library of congress subjects (subjects_fixed.jsonl), and the 'name' field from the opensyllabus data. This text data is then converted into embeddings using the [bge-large-en-v1.5](https://huggingface.co/BAAI/bge-large-en-v1.5) model.

The embeddings are clustered into topics using K-means clustering. We include data for k=10,000 (all_topic_clusters_10000.parquet) and k=20,000 (all_topic_clusters_20000.parquet). We also include the centroid embeddings for each cluster. Both runs used 100 iterations of the k-means algorithm.