This code is for Bestbuy - Predict which Xbox game a visitor will be most interested in based on their search query (https://www.kaggle.com/c/acm-sf-chapter-hackathon-small).
The approach is pretty straighforward, the basic idea is try to match a query with a game (so I ignore all other features like time, game name etc.). The initial approach is to create a map, the key is user's query, the value is also a map<game, count>, which means I store which game the user clicked and the frequecy of it when user searches for this query. After that, when predicting the game, we simply choose the most frequently clicked games for this query, and if there's less than 5 games, I use the most popular games among all queries to fill the gap.
One optimization I used is to correct users' queries. Since there are lots of typos or short forms, I use Levenshtein to calculate two queries similarity, and if it is above a threshold, the algorithm believe they are the same query. The threshold is chosen by testing on cross validation set.
Finally, I got around 74.3% and the leader is 78.9%.