I was solving a problem to compare 3 million, 2018 documents against the 2019 documents. There are three text attributes to be compared from one item against the other. I used Latent Semantic Indexing (LSI) for one variable, containing about 5 word tokens, with reasonable performance.
- What will be the minimum document size for LSI/LDA for a multivariate [Item Description (text) (10 tokens), Item specification (text) (5 tokens)] problem to compute document similarity?
- I had used cosine similarity, String Distances to measure the closeness of match of 2018 Description and 2019 description. Are there any other statistical methods to evaluate the Model?