If the similarity is very high then when users go through item A, at the same time we can recommend item B. Because these properties are not dependent on users, but it is suitable for a small amount of data ,and thus it is not good for personal recommendations. The idea of this algorithm is clear, but the simpler algorithm is, the more difficult the get a better result. Especially the conversion of rate is very low. Products may have dozens of commercial properties, but there is no necessary to calculate all properties of product .For those which have been included in properties, the importance is different from one to another.So if we do every product by using this method, the calculation must be very huge . Meanwhile, we have to consider
…show more content…
For example, the user A bought products 1, 2, 3, 4, the user B users bought products 1, 2, 5, 6, you can simply 1234 and 1256 respectively as attribute string for user A and user B, then we calculate A and B’s similarity. After calculations, users will be clustered into several categories. Assuming that B and A belongs to a cluster, then it can say that A and B has similar preferences, and if A bought it but B does not, we can recommend it to B. In this process, we need to consider: 1.categorize user (such as, the number of purchased items ,or timeless of the data span and so on); 2. calculate the similarity when mentioned on the first aspect, because users are not the same. It is possible that the importance of low price product is less important then high price products. 3, cluster categories, the clustering algorithm is another subject, or choose a classification algorithm. The clustering takes a long time to test, observe, modify, and accumulate. 4, estimate the viewed history , purchase, and collection if they can do collaborative
For example, the user A bought products 1, 2, 3, 4, the user B users bought products 1, 2, 5, 6, you can simply 1234 and 1256 respectively as attribute string for user A and user B, then we calculate A and B’s similarity. After calculations, users will be clustered into several categories. Assuming that B and A belongs to a cluster, then it can say that A and B has similar preferences, and if A bought it but B does not, we can recommend it to B. In this process, we need to consider: 1.categorize user (such as, the number of purchased items ,or timeless of the data span and so on); 2. calculate the similarity when mentioned on the first aspect, because users are not the same. It is possible that the importance of low price product is less important then high price products. 3, cluster categories, the clustering algorithm is another subject, or choose a classification algorithm. The clustering takes a long time to test, observe, modify, and accumulate. 4, estimate the viewed history , purchase, and collection if they can do collaborative