We are going to discuss a bad example of recommendation agent (RA) used by an Indonesian e-commerce/ online forum platform, Kaskus. You would immediately think that this platform performs badly these days, but you cannot be more wrong. Recently, Kaskus announced that the site has achieved 600,000,000 page views every month and 40,000,000 users registered to the site which certify Kaskus to be the biggest online forum in Indonesia (Lukman, 2014) and Indonesia’s #1 website in 2013 (Redwing, 2013).
Kaskus started as an online bulletin board forum for gamer communities (Wee, 2012). When the traffic picked up, the users saw the opportunity to sell their things in the forum and thus e-commerce threads were popping up everywhere. Seeing how many e-commerce threads were created, Kaskus established a sub-forum called Kaskus Jual Beli (KJB) or Kaskus Selling and Buying especially for e-commerce threads.
For a long period of time, KJB has no recommendation agent what so ever to assist the buyers when purchasing something. What it had was merely content filtering system which was very inefficient because KJB had too many thread posts. Recently, however, KJB started to implement rule based preference elicitation to its existing RA system. This new type of RA is arguably effective because there are many different types of item sold. For example, if we want to buy a puppy (Yes, real puppy!) then the condition rule such as “new” or “second” will not be valid anymore.