Doubt on scope of text classification problem

Question

I have a dataset that describes the sellers who are selling various brands. I need to identify the source (where did he buy those brands he is selling from) of those sellers. (Dimension of dataset 11,29,490 rows and 2 columns: Seller and brand)

eg:

I need unique instances for each seller. That is, in each row, I have to have a seller and all the information about the brands he is selling. My idea was to create make each unique brand as a feature and create a sparse dataset. However, we have almost 2 million unique brands. Is it possible to do some text classification and come up with some text clusters and then make each cluster as a feature? I am not sure if this is a correct approach I am stuck without an approach now. Can anyone help me with how should I approach this problem?

Thanks in advance

Answer 1

In case you have a lot of cases where the same brand is described in different ways, e.g. "K.F.C", "KFC", "Kentucky Fried Chicken", then it might be worth using clusters based on textual similarity measures. This kind of problem is similar to record linkage.

But if this is not the case, then it would be a bad idea to try to merge brands based on their names since you would end up with groups of brands which have nothing in common except some part of their name.

At a more general level, it looks to me like the design of your task is flawed: why do you need a unique instance for every seller? A seller can have multiple brands, each from a different source, right? That would mean that there are multiple sources for each seller, how do you plan to deal with that? Are you sure you don't want to predict the source for a particular pair seller + brand instead?