Using temporal IDF for efficient novelty detection in text streams (1401.1456v2)

Abstract: Novelty detection in text streams is a challenging task that emerges in quite a few different scenarios, ranging from email thread filtering to RSS news feed recommendation on a smartphone. An efficient novelty detection algorithm can save the user a great deal of time and resources when browsing through relevant yet usually previously-seen content. Most of the recent research on detection of novel documents in text streams has been building upon either geometric distances or distributional similarities, with the former typically performing better but being much slower due to the need of comparing an incoming document with all the previously-seen ones. In this paper, we propose a new approach to novelty detection in text streams. We describe a resource-aware mechanism that is able to handle massive text streams such as the ones present today thanks to the burst of social media and the emergence of the Web as the main source of information. We capitalize on the historical Inverse Document Frequency (IDF) that was known for capturing well term specificity and we show that it can be used successfully at the document level as a measure of document novelty. This enables us to avoid similarity comparisons with previous documents in the text stream, thus scaling better and leading to faster execution times. Moreover, as the collection of documents evolves over time, we use a temporal variant of IDF not only to maintain an efficient representation of what has already been seen but also to decay the document frequencies as the time goes by. We evaluate the performance of the proposed approach on a real-world news articles dataset created for this task. The results show that the proposed method outperforms all of the baselines while managing to operate efficiently in terms of time complexity and memory usage, which are of great importance in a mobile setting scenario.