Are Recommendation Engines a Threat to the Long Tail? Yes, but isn’t that the point?
Read/Write Web posted and article a couple weeks back about a study done by Two Wharton academics that asks “whether online recommendation services are a threat to the aggregate diversity of items discovered by their users. The study is titled “Blockbuster Culture’s Next Rise or Fall: The Impact of Recommender Systems on Sales Diversity” . The argument boils down to whether these systems help people discover only popular content or truely new content.
I believe that the point of a recommendation service to help retailers or marketers move inventory. People already know and desire the popular product (head of the tail), because it is in the social consciousness. There isn’t a need to recommend items that may have been just released. The manufacturer/publisher/content creater is already running some type of campaign to get a consumers attention. Instead, a recommender system should be putting items in front of a user that may have gone out of the social consciousness. Say for example a movie that came out 6 months ago in the theatre and is now being released on video, or a book that isn’t on some top 10 list anymore or even re-runs of tv shows. These are the items that you want to show users. They most certainly aren’t new but they may be new to the user/consumer.
There is a third part of the recommendation system that I think people and some companies are missing, and that is the conversation with the user. Lets take movie recommendations as an example, when you start to talk to your friends about movies you sometimes say “You should see this movie it is great.” After that initial sentence the conversation usually boils down to have you seen this movie. did you like it? Subconsciously you are building a profile of your friend about what movies they like so that you can recommend them or not. Recommendation systems should do the same thing. It should ask the users about movies that they may have seen or might like to see. People can’t remember all the movies that they have seen, but if you ask them about movies more than likely they have seen a good number of them. This does a couple of things, first it helps the system build a more accurate profile of the user instead of relying on purchase history or rating random things. Secondly it gets the user involved, they like to give their opinions on things. Lastly it helps get ratings for those long tail items and helps with the proverbial “cold start” problem that one always hear about when talking about recommender systems.
If you want to see this at work check out the Sourcelight Discovery Guide. There you can see the conversation that is implemented. As you get more experience with the system it will attempt to go deeper into the long tail and have a conversation about those items. It also attempts to show you items that it thinks you have seen. So if you rate Indiana Jones it is going to give you all the Indian Jones movies to rate, because you most likely have already seen then and might enjoy rating them.
Are recommender systems a threat to the long tail? Yes I believe that they are, but they will never eliminate the long tail, just flatten it out some. Perhaps when recommender systems get good enough we will have some type of double humped demand graph (camel tail) where initial demand is high then demand is high again in the future then it tails off. People will always want to do what their social contacts do, initially they all want to consume the same things, in private those seem people desire to consume individual things and often do. The job of a recommender system is to help people discover items to buy/rent helping your clients achieve their financial goals.
Share This-->