arianpasquali/rollcall-factbase-trump-discourses

Factbase Trump Discourses (June 2015 — February 2026)

Full-text transcripts of 3,925 public communications by Donald Trump, spanning his first presidential campaign through his second term. Sourced from Factbase.

Dataset Description

Each record is a single document (speech, interview, press conference, etc.) with its full transcript and metadata. The collection covers over a decade of political discourse across 12 document types.

Document Types

Type	Count
Remarks	1,455
Speech	844
Interview	758
Press Gaggle	405
Donald Trump Vlog	195
Press Conference	173
Weekly Address	59
Debate	18
Convention	11
Leaked Remarks	5
Prepared Remarks	1
Deposition	1

Coverage

• Date range: June 16, 2015 — February 9, 2026
• Total documents: 3,925
• Total text: 112M characters (28,700 characters per document on average)
• Shortest document: 24 characters
• Longest document: 225,750 characters

Schema

Field	Type	Description
filename	string	Original source PDF filename from Factbase
title	string	Document title
file_size_kb	float	Size of the original source PDF in KB
page_count	int	Number of pages in the original PDF
type	string	Document type (Speech, Remarks, Interview, etc.)
transcript	string	Full text transcript
speech_date	string	ISO 8601 timestamp of the speech/event

Example Record

{
  "filename": "Remarks_ Donald Trump Listening Session with Airline Executives - February 9, 2017.pdf",
  "title": "Remarks: Donald Trump Listening Session with Airline Executives",
  "file_size_kb": 537.34,
  "page_count": 24,
  "type": "Remarks",
  "speech_date": "2017-02-09T00:00:00.000",
  "transcript": "..."
}

Usage

from datasets import load_dataset

dataset = load_dataset("arianpasquali/rollcall-factbase-trump-discourses")
df = dataset["train"].to_pandas()

# Filter by type
speeches = df[df["type"] == "Speech"]

# Filter by date range
df["date"] = pd.to_datetime(df["speech_date"])
covid_era = df[(df["date"] >= "2020-03-01") & (df["date"] <= "2021-01-20")]

Use Cases

• Discourse and rhetoric analysis — study rhetorical patterns, framing, and narrative strategies across different contexts and time periods
• Emotion and sentiment classification — train or evaluate emotion/sentiment models on political speech
• Topic modeling — identify recurring themes and how they evolve over time
• NLP benchmarks — long-form political text for summarization, information extraction, or question answering tasks
• Temporal analysis — track linguistic and thematic shifts across campaigns, presidency, and post-presidency

Source

Transcripts were collected from Factbase, a nonpartisan archive of presidential communications. The original PDF transcripts were parsed and structured into this JSON dataset.

License

This dataset is released under CC BY-NC 4.0. The underlying transcripts are public-domain government records and third-party interview transcripts aggregated by Factbase.