Finding Similar Documents
Once you’ve queried and discovered documents of interest, one of the next things you might want to do is find similar documents. Whereas TF-IDF can provide the means to narrow down a corpus based on search terms, cosine similarity is one of the most common techniques for comparing documents to one another, which is the essence of finding a similar document. An understanding of cosine similarity requires a brief introduction to vector space models, which is the topic of the next section.
The Theory Behind Vector Space Models and Cosine Similarity
While it has been emphasized that TF-IDF models documents as unordered collections of words, another convenient way to model documents is with a model called a vector space. The basic theory behind a vector space model is that you have a large multidimensional space that contains one vector for each document, and the distance between any two vectors indicates the similarity of the corresponding documents. One of the most beautiful things about vector space models is that you can also represent a query as a vector and find the most relevant documents for the query by finding the document vectors with the shortest distance to the query vector. Although it’s virtually impossible to do this subject justice in a short section, it’s important to have a basic understanding of vector space models if you have any interest at all in text mining or the IR field. If you’re not interested in the background theory and want to jump ...