Twitter bots

• Exercise in: adapting existing code

  • Time to market: don't need to handle irrelevant parts of the Twitter api myself
  • Resiliance: If Twitter API changes, I only need to update the library and not my code

• First two bots: comparison bots

• Functional decomposition

  • Load tweets
  • Extract word/adjective pairs
  • Find adjectives with more than two nouns/nouns with more than two adjectives
  • Generate message containing a random adjective/noun from this list
  • Post tweets with said message

• The first three parts:

  • Takes a long time
  • only needs to be done once

• So separate into two stages: preprocess archive and then have a lightweight bot that only runs the last steps

• No reference design for word pairs: researched solutions with similar aim: grammar checkers -> link grammar

• Wrote a Java program to extract

• Bot: reference design: Misp's twitter_ebooks bots

• Bot is written in simple Ruby based on example bot

• Second two bots: image processing bots

• Again, fn. decomp for modularity

• Listens for pictures sent to account

• Detect face and Add images based on face detection result

• Send new picture as a reply

• Reference design: Android's face detection ability

• Found a port of Android's library to normal Linux

• Fn. Decomp lets me work on this part separately from the bot that listens to replies

• Before bot part was done, to build a medium-fidelity prototype, wrote a script that loads files from local hard drive into image generator so I can easily tweak image positions

• when done, integrate work with twitter bot

• When adding new bot, realized that bot was not modular enough:

  • detect face - doesn't change; add images - changes

• Refactored code to separate face detection and overlay logic for reuse

  • Iteration