Reputation: 1065
I am in the process of learning Singular Value Decomposition and for what purposes I can use this concept and the book that I am reading mentioned that SVD is used in Latent Semantic Indexing. I read few articles about LSI and it seems like LSI is mainly used in search engines and in similar applications. I wanted to use LSI for a small data analysis project that I am working on and I am not sure if it makes sense for my application. Here is what I am working with.
I have a list of about 20000 games and 2 of the attributes of this list are the game genre and the platform the game was released on. I wanted to use LSI to gain some information about the platform and the genre attributes.
So at first I created a Co-occurrence matrix where rows represent the 24 different genres and columns represent 22 different platforms. Then I did SVD decomposition of the co-occurrence matrix and extracted first two columns of U and V and made a 2d plot for U and V.The plot looks like this.
So my question is, can this be considered a meaningful usage of Latent Semantic Indexing and also how can I interpret from this graph ? For example, we see that Genre Action and platform PC are far away from all other variables, does this tell us anything about this genre and platform in particular ?
Thank you.
Upvotes: 2
Views: 697
Reputation: 727
SVD is usually used to reduce dimensionality and uncover hidden macro characteristics or pattern for a more fine grained behaviour. Almost all recommendation system problem could be approached with this method.
I am not sure about the parameter you used to used to realise your SVD (if it has really converged) but some possible interpretations of your graph :
good luck
Upvotes: 0