🌲Tree-of-Text

"Tree-of-Text: A Tree-based Prompting Framework for Table-to-Text Generation in the Sports Domain", ACL SRW 2025 Long Paper (Oral)

Paper | Slides | Demo

Setup

1. Create the conda environment

   conda env create -f environment.yml

2. Activate the conda environment

   conda activate Tree-of-Text

Data

ShuttleSet+ (40/9/9)

• table
  • match.csv: overall information for the match
  • set_1.csv: rally-level data for set 1
  • set_2.csv: rally-level data for set 2
  • set_3.csv: rally-level data for set 3
• text
  • text.txt: human-written report for the match from online sources such as the BWF and Olympics websites

RotoWire-FG (5340/1147/1148)

• table
  • game.csv: overall game information
  • home_line.csv: line score of the home team
  • vis_line.csv: line score of the visiting team
  • box_score.csv: individual player statistics
• text
  • text.txt: human-written NBA basketball game summary

MLB (22821/1739/1744)

• table
  • game.csv: overall game information
  • home_line.csv: line score of the home team
  • vis_line.csv: line score of the visiting team
  • box_score.csv: individual player statistics
  • play_by_play.csv: scoring of each at-bat
• text
  • text.txt: human-written baseball summaries sourced from the ESPN website

Code

ShuttleSet+

• Tree-of-Text.ipynb: follow the instructions in the script and execute the steps in order: Config, Generate Report, Relation Extraction, Evaluate Report.
• prompt
  • planning: prompt for Content Planning
  • write: prompt for write() operation
  • generating: prompt for Content Generating

RotoWire-FG

• Tree-of-Text.ipynb: follow the instructions in the script and execute the steps in order: Config, Generate Report, Relation Extraction, Evaluate Report.
• prompt
  • planning: prompt for Content Planning
  • write: prompt for write() operation
  • generating: prompt for Content Generating

MLB

• Tree-of-Text.ipynb: follow the instructions in the script and execute the steps in order: Config, Generate Report, Relation Extraction, Evaluate Report.
• prompt
  • planning: prompt for Content Planning
  • write: prompt for write() operation
  • generating: prompt for Content Generating

Demo

ShuttleSet+

1. Configuration: Framework hyperparameters · Model settings
2. Input Tables: ShuttleSet+ dataset · Three tables
3. Tree-of-Text: Click a node to view its details · Filter by operation type
4. Output Texts: Reference Text vs. Generated Text

RotoWire-FG

1. Configuration: Framework hyperparameters · Model settings
2. Input Tables: RotoWire-FG dataset · Four tables
3. Tree-of-Text: Click a node to view its details · Filter by operation type
4. Output Texts: Reference Text vs. Generated Text

MLB

1. Configuration: Framework hyperparameters · Model settings
2. Input Tables: MLB dataset · Five tables
3. Tree-of-Text: Click a node to view its details · Filter by operation type
4. Output Texts: Reference Text vs. Generated Text