Temporal analysis of text data using latent variable models

Abstract

Detecting and tracking of temporal data is an important task in multiple applications. In this paper we study temporal text mining methods for Music Information Retrieval. We compare two ways of detecting the temporal latent semantics of a corpus extracted from Wikipedia, using a stepwise Probabilistic Latent Semantic Analysis (PLSA) approach and a global multiway PLSA method. The analysis indicates that the global analysis method is able to identify relevant trends which are difficult to get using a step-by-step approach. Furthermore we show that inspection of PLSA models with different number of factors may reveal the stability of temporal clusters making it possible to choose the relevant number of factors.

Discovering Temporal Patterns from Images using Extended PLSA

“…[5] conducted experiments on a synthesize image data set and a real news data set, 20NG data set, consisting of approximately 20,000 news group documents and also on the Tsunami news data set used by [2]. [6] investigated the description of composers in Wikipedia by retrieving all documents that Wikipedians assigned to composer categories such as "Baroque composers" and "American composers", result in a collection of 7358 documents. [7] also considered the database of NIPS papers as well as the US Presidential State of the Union address.…”

“…As [2] and [4] do. [6] adopted PLSA model to extract topics from each sub-collection, and then stringed together for measuring similarity between each other using the KLdivergence.…”

“…[5] used an online Nonnegative Matrix Factorization(ONMF) algorithm to detect latent factors and track their evolution while the data evolve by leveraging the already detected latent factors and the newly arriving data. Except the stepwise PLSA model, [6] also present another global multiway PLSA method to consider the temporal element holistically. The Topic over Time(TOT) model presented in [1] modelled the distribution over topic considering both word co-occurrences and the document's time stamp together.…”