MLB Robot Jounralism

About

This is a program that fetches play-by-play data of a game from MLB.com and analyzes the overall trend of the game. Using the analysis, it detects the highlights of the game and generates a short article (in Korean) on the game chosen.

Authors

• Written by Sean D Kim from Carnegie Mellon University
• Under the guidance of Professor Joon-Hwan Lee and Dong-Hwan Kim from Seoul National University

Language/Module

• Ruby on Rails
• Nokogiri for parsing xml file

Data Source

• All the game data from MLB.com
• WE (Win Expectany) and WPA (Win Percentage Added) is calcuated using table provided by fangraphs.com (?) This table can be found in winexp.csv

Files

• teamnames.csv: contains abbreviation and the code name of each team
• winexp.csv: contains WE of a specific situation of a baseball game

Method

The program consist of three main parts.

1. Data scraping

• All the MLB game data are organized at http://www.mlb.com/gdcross/components/game/mlb/
• To find a specfic game data, navigate through directories in the following order.
  • year => month => day => game (starts with gid_...)
• Once specific data is found, the following info can be found in the following folders
  • players.xml: list of player name and their 6 digit number ID (this number ID is used in the game data)
  • game_events.xml: every event in the game is recorded here.

2. Data Analysis

• From game_events.xml:
  • At every element, the following data was recorded from the following attributes
    • Away team score from attribute "away_team_runs"
    • Home team score from attribute "home_team_runs"
    • Current Inning from the parent element,
    • Ball count from attribute "b"
    • Strike count from attribute "s"
    • Out count from attribute "o"
    • Player present in the first base from attribute "b1"
    • Player present in the second base from attribute "b2"
    • Player present in the third base from attribute "b3"
    • Current pitcher from attribute "pitcher"
    • Current batter from attribute "batter"
    • Type of event from attribute "event" (ex. flyout, single)
    • Description of the event from attribute "des"
  • At every element, update the necessary situation change using adequate attributes.
    • Stolen Base: Attribute "event" indicates which base was stolen. (ex. event="Stolen Base 2b") Update the base situation accordingly.
    • Wild Pitch: Attribute "des" identifies the base situation changes. Change accordingly.
    • Game Advisory, Manager Review, Pitching Substitution, Offensive Sub, Defensive Sub: Do nothing.
    • Pickoff Error 1B: (Yet Implemented)
• Additionally, WE (Win Expectany) and WPA (Win Percentage Added) was calculated using winexp.csv
  • calculate column and row value
    • column: (away team points) – (home team points)
    • row: four digit number generated in the following way
      • 1st digit: Current inning
      • 2nd digit: 1 if Top (away team on offense), 2 if Bottom (home team on offense)
      • 3rd digit: if the player(s) are at the following base(s)...
        • none => 1
        • first base => 2
        • second base=> 3
        • first & second base => 4
        • third base=> 5
        • first & third base => 6
        • second & third base => 7
        • all bases => 8
      • 4th digit: number of outs
  • the WE is the number found at the column and row calculated above
  • WPA is calculated by subtracting previous WE from the current WE
  • WE is closer to 1 if home team is more likely to win, closer to 0 if away team is more likely to win

3. Generating Article

• The highlights of the game is identified by highest change in WE (greatest |WPA|)
• The sentence is generated using algorithm developed by the research team of Professor Joon-Hwan Lee from Seoul National University

Shortcomings

The followings are the shortcomings that needs to be fixed.

• elements
  • not all events are being taken care of
    • Pickoff: base situation update necessary
    • Substitution: change the player
    • Search for more events that could happen in a element
  • when the first element is , the pitcher/batter is empty
• Database
  • Current implementation does not make usage of database. Loading players and storing game data can be done more effectively using database.