sql server - Recommendation system experimentation -

I am reading book programming collective intelligence and trying to apply what I have learned against the Northwind database. Although I have not yet convinced in my understanding of the introduction of algorithms, I am starting to get the views of common concepts.

1. By using the Northwind database, I am trying to show a list of "bought XYZ bought" by using the following pretenders
   1. Search for other customers
   2. Rank items of items based on the purchase
   3. Return the top n items from the previous step

   I am working with the question given below:

     Deactivate @ Customerless (5), @ Producted Ent; Set @customerid = 'ALFKI'; Set @producted = 59; - Find other products from those customers - who have also bought my producer, in the top order, from 10 od.productid, c.categoryname, p.productname, p.unitsonorder, count (od.productid) [order details] od inner inner Order o o.orderid = od.orderid internal included product p.productid = od.productid in internal categories c c categorid = p.categoryid where o.customerid & lt; & Gt; @customed and odd.product & lt; & Gt; Group by @productid and p.discontinued = 0 od.productid, c.categoryname, p.productname, p.unitsonorder command 5 characters, 4 desc  

   I think my next step To break the query so that I can filter recently based on purchases (as opposed to all historical purchases) and match the customer compared to all the customers purchased by my customer. Can anybody provide any signal? Am I going in the right direction? Should I take a completely different direction?

   At this point, my goal is to display accuracy, because I know that I do not yet have the experience of implementing algorithms for their full benefit. I'm just trying to implement the concept. Once I am satisfied, I understand it well, I want to test this query against a larger database with more realistic customer data.

   text ">

   It is commonly called" market basket "analysis or" affinity group "and it is not as simple as it seems, due to most combinations. The first thing is that the minimum of cases For example, suppose we have 1 M baskets and 10 people who had bought the product B in the same basket as the product A had bought. In 10 cases, It is worth noting that "those who have bought it have also bought B". For this reason, using "recent" purchases is a bit difficult. This idea is to create a table of product pairs, but if we have N Products, The number is N * (N-1), so for 1000 product stores, we have 999000 combinations, so the algorithm which is not it is simple.

   The order of objects to consider one more thing And the value of each is for example cy Customers who purchase only can buy a "LED light" (add) But, if someone puts a LED light in a basket, then can the system present a bicycle in the list?

   In this meditation Keeping that you are using SQL Server, I use the "MS Association Algorithm" for this purpose to you , using the same data, you have your results against "business solution" The Lna can.