Dynamic topic modeling is useful at discovering the development and change in latent topics over time. However, present methodology relies on algorithms that separate document and word representations. This prevents the creation of a meaningful embedding space where changes in word usage and documents can be directly analyzed in a temporal context. This paper proposes an expansion of the compass-aligned temporal Word2Vec methodology into dynamic topic modeling. Such a method allows for the direct comparison of word and document embeddings across time in dynamic topics. This enables the creation of visualizations that incorporate temporal word embeddings within the context of documents into topic visualizations. In experiments against the current state-of-the-art, our proposed method demonstrates overall competitive performance in topic relevancy and diversity across temporal datasets of varying size. Simultaneously, it provides insightful visualizations focused on temporal word embeddings while maintaining the insights provided by global topic evolution, advancing our understanding of how topics evolve over time.

Removing underground, unexploded bombs, mortars, cannon-shells and other ordnance ("MEC" or "UXO") from former military ranges is difficult and expensive. The principal difficulty is discriminating intact, underground ordnance from other metallic items such as fragments of exploded ordnance ("Clutter"), magnetic rocks, and "historic" items such as horseshoes, barbed-wire, and refrigerators. This study represents the first, large-scale, blind-test of MEC discrimination technology on production-grade, survey-mode data from the cleanup of a real impact site. The results reported here significantly advance the state-of-the-art in MEC discrimination over alternative forward-modeling/inversion approaches to performing MEC discrimination. We combined Linear Genetic Programming (LGP) and statistical analysis to process data from the cleanup of 600 acres of the F.E.Warren Air Force Base. That data contained almost 30,000 targets of interest identified by geophysicists, including three-hundred thirty-two 75mm projectiles (75mm) and 37mm projectiles (37mm). A little under one-third of the groundtruth was held back by the customer for blind-testing. Our task was to discriminate intact 37mm's and 75mm's from the clutter by ordering the targets from most-likely to be MEC to least-likely to be MEC in what is referred to as a "prioritized dig-list". We identified all 75mm's by 28.2% of the way through our prioritized dig-list and all 37mm's by 64.2% of the way through the prioritized dig list. Thus, depending on ordnance type, we reduced the number of targets that had to be excavated (false alarms) to clear the entire site by between 35% and 72%.