Google Feudler: An autocomplete guessing game

Spring 2017 — Web app

This project was the final group project for Brown CS’s software engineering course. Our goal was to create a web app of the game Google Feud, a spinoff of Family Feud. Users are given an incomplete Google query, and they guess how Google would complete the query. Users are awarded points according to how high up on Google’s rankings words are. The game supports multiplayer functionality, as well as categories that can be chosen.

This game already exists, yet there was a major drawback to putting the game on the web. The original game involves a person reading the query, and comparing guesses with the real suggestions on Google. This allows the game to be more playable — if users guessed words that were close in meaning to the suggestions, then they would be awarded points. In the existing web version, only exact matches were awarded points. From initial game testing, we quickly realized that finding exact matches for most queries was very difficult, and in most rounds players only guessed 0-1 suggestions correctly. This caused much frustration among players, which is why we decided to recreate the game addressing this issue.

I worked on comparing the semantics of guesses with the semantics of suggestions, in order to award points to guesses that are close enough in meaning to a suggestion. The core of the approach built off of Word2Vec, a model that converts words to vectors that contain semantic information. We used cosine similarity to score similarity of words, and used a handmade dataset to threshold

Another feature we added to the base game is what we called “meta-mode”. We collected data about how users guess words for input queries. Fundamentally, guessing what people would search is different than actually searching, and so we created a mode where the answers are the most guessed words instead of Google’s suggestions.

The code for this project is on GitHub.