"Temporal Latent Personal Analysis (T-LPA)"

Domain-based Latent Personal Analysis (LPA) is an unsupervised method for spectral exploration of entities such as users' styles, documents, and more. Its power lies in the creation of a domain, a collection, and finding for each document a signature that captures how the document differs from the domain. A document's signature contains popular domain terms underused in the document and terms overused in the document.

Here, we expand LPA to consider temporal changes in what we term T-LPA. T-LPA detects temporal changes in the general frequency of terms, taking into account their age, and calculates their relative importance given their current and previous prevalence. We apply T-LPA to a large corpus containing both mainstream, non-conspiratorial, and conspiratorial stories published during the first year of the Covid-19 pandemic and follow the trail of conspiratorial narratives. Our results show that conspiratorial narratives find their way to mainstream media with time.

Bio: Osnat (Ossi) Mokryn is the Social Content and Networks (SCAN) lab director and a senior lecturer at the Information System department at the University of Haifa. Her main research areas are computational social science and temporal networks. In her research, she applies multidisciplinary methods to identify governing principles that facilitate the systematic analysis of highly complex systems. Recently, her research has focused on dividing methods for spectral user modeling, contextual recommender systems, and the temporal evolution of networks.

Ossi holds a Ph.D. in Computer Science from the Hebrew University in Jerusalem, an M.Sc. in Electrical Engineering, and a B.Sc. in Computer Engineering from the Technion, Israel Institute of Technology.