Embeddings of Item 1 from 10-K Filings (1994-2022) using MPNet

This dataset contains sentence embeddings generated from Item 1 (Business Description) of 10-K filings submitted to the SEC between 1994 and 2022. The filings' headers were obtained from the SEC EDGAR database via WRDS (Wharton Research Data Services). The text from each Item 1 was extracted, cleaned to remove tables, images, headers, footers and comments, and then tokenized into sentences using NLTK's Punkt tokenizer. The sentences were then embedded using the all-MPNet-base-v2 model from Sentence Transformers, and the embeddings for each filing were averaged to create a single embedding vector per filing. The dataset includes the following fields for each filing: file_id: A unique identifier for the filing embedding: The averaged sentence embedding for the Item 1 text n_words: The number of words in the Item 1 text This dataset was created for a research paper examining the impact of business description similarity on analyst coverage and investment recommendations. It allows similarity comparisons between companies based on the semantic content of their business descriptions. An additional index file in CSV format is provided to map the file_id to the following filing metadata: fdate: Filing date cik: Central Index Key (CIK) of the filing company url: URL of the filing on the SEC EDGAR database