TDSM 1.13

There are two general methods.

1. Content based recommender systems: These system believe that when a person is looking for specific items, it is probable that he likes the similar products. So based on the features of the product we can define the similarity. What is important here is defining the similarity, and defining good features for our products in the model. One problem here would be not necessarily all people who searches for a product are interested in it.

2. Collaborative filtering: These systems believe that two similar users look for similar products. So here the problem will be defining features for users and a metric to compute similarity between people. Here the problem would be we need enough information from people to build a model based on them.

Building a recommender system contains these steps:

1. Extracting features: To extract the features for each user, we should answer these questions: Which products are more popular? What products each user likes more? Is time slot important in popularity of the feature?

2. Building model based on the features: We should model the features to predict P(I|U) where U is the user and I is the item. We can use Baysian theorem to build classifier that predict which items are more popular for each user.

3. Improving model: We should modify our model over time based on the feed back of the users.