Jean Mark Gawron

Mapping social activities and concepts with social media (Twitter) and web search engines (Yahoo and Bing): a case study in 2012 US Presidential Election

We introduce a new research framework for analyzing the spatial distribution of web pages and social media (Twitter) messages with related contents, called Visualizing Information Space in Ontological Networks (VISION). This innovative method can facilitate the tracking of ideas and social events disseminated in cyberspace from a spatial-temporal perspective. Thousands of web pages and millions of tweets associated with the same keywords were converted into visualization maps using commercial web search engines (Yahoo application programming interface (API) and Bing API), a social media search engine (Twitter APIs), Internet Protocol (IP) geolocation methods, and Geographic Information Systems (GIS) functions (e.g., kernel density and raster-based map algebra methods). We found that comparing multiple web information landscapes with different keywords or different dates can reveal important spatial patterns and “geospatial fingerprints” for selected keywords. We used the 2012 US Presidential Election candidates as our case study to validate this method. We noticed that the weekly changes of the geographic probability of hosting “Barack Obama” or “Mitt Romney” web pages are highly related to certain major campaign events. Both attention levels and the content of the tweets were deeply impacted by Hurricane Sandy. This new approach may provide a new research direction for studying human thought, human behaviors, and social activities quantitatively.

Practical Issues in Compiling Typed Unification Grammars for Speech Recognition

Current alternatives for language modeling are statistical techniques based on large amounts of training data, and hand-crafted context-free or finite-state grammars that are difficult to build and maintain. One way to address the problems of the grammar-based approach is to compile recognition grammars from grammars written in a more expressive formalism. While theoretically straight-forward, the compilation process can exceed memory and time bounds, and might not always result in accurate and efficient speech recognition. We will describe and evaluate two approaches to this compilation problem. We will also describe and evaluate additional techniques to reduce the structural ambiguity of the language model.

Mapping ideas from cyberspace to realspace: visualizing the spatial context of keywords from web page search results

We introduce a new method for visualizing and analyzing information landscapes of ideas and events posted on public web pages through customized web-search engines and keywords. This research integrates GIScience and web-search engines to track and analyze public web pages and their web contents with associated spatial relationships. Web pages searched by clusters of keywords were mapped with real-world coordinates (by geolocating their Internet Protocol addresses). The resulting maps represent web information landscapes consisting of hundreds of populated web pages searched by selected keywords. By creating a Spatial Web Automatic Reasoning and Mapping System prototype, researchers can visualize the spread of web pages associated with specific keywords, concepts, ideas, or news over time and space. These maps may reveal important spatial relationships and spatial context associated with selected keywords. This approach may provide a new research direction for geographers to study the diffusion of human thought and ideas. A better understanding of the spatial and temporal dynamics of the ‘collective thinking of human beings’ over the Internet may help us understand various innovation diffusion processes, human behaviors, and social movements around the world.

Social media analytics and research testbed (SMART): Exploring spatiotemporal patterns of human dynamics with geo-targeted social media messages

The multilevel model of meme diffusion conceptualizes how mediated messages diffuse over time and space. As a pilot application of implementing the meme diffusion, we developed the social media analytics and research testbed to monitor Twitter messages and track the diffusion of information in and across different cities and geographic regions. Social media analytics and research testbed is an online geo-targeted search and analytics tool, including an automatic data processing procedure at the backend and an interactive frontend user interface. Social media analytics and research testbed is initially designed to facilitate (1) searching and geo-locating tweet topics and terms in different cities and geographic regions; (2) filtering noise from raw data (such as removing redundant retweets and using machine learning methods to improve precision); (3) analyzing social media data from a spatiotemporal perspective; and (4) visualizing social media data in diagnostic ways (such as weekly and monthly trends, trend maps, top media, top retweets, top mentions, or top hashtags). Social media analytics and research testbed provides researchers and domain experts with a tool that can efficiently facilitate the refinement, formalization, and testing of research hypotheses or questions. Three case studies (flu outbreaks, Ebola epidemic, and marijuana legalization) are introduced to illustrate how the predictions of meme diffusion can be examined and to demonstrate the potentials and key functions of social media analytics and research testbed.

Social media analytics and research test-bed (SMART dashboard)

We developed a social media analytics and research testbed (SMART) dashboard for monitoring Twitter messages and tracking the diffusion of information in different cities. SMART dashboard is an online geo-targeted search and analytics tool, including an automatic data processing procedure to help researchers to 1) search tweets in different cities; 2) filter noise (such as removing redundant retweets and using machine learning methods to improve precision); 3) analyze social media data from a spatiotemporal perspective, and 4) visualize social media data in various ways (such as weekly and monthly trends, top URLs, top retweets, top mentions, or top hashtags). By monitoring social messages in geo-targeted cities, we hope that SMART dashboard can assist researchers investigate and monitor various topics, such as flu outbreaks, drug abuse, and Ebola epidemics at the municipal level.

Topic Models: A Tutorial with R

This tutorial presents topic models for organizing and comparing documents. The technique and corresponding discussion focuses on analysis of short text documents, particularly micro-blogs. However, the base topic model and R implementation are generally applicable to text analytics of document databases.

Mapping Spatial Information Landscape in Cyberspace with Social Media

This chapter describesa Spatial Web Automatic Reasoning and Mapping System (SWARMS) for visualizing and analyzing space-time dimensions of information landscape represented by a social media channel—Twitter. SWARMS utilizes computer programming and Twitter Search APIs to retrieve tweets by searching keywords from the Twitter database. Two case studies were conducted to analyze the spatial information landscape: the 2012 U.S. Presidential Election and 2012 summer movies. The two case studies were selected because these events can have a reality check by comparing to the actual election results and the movie box office revenue. Our preliminary spatial analysis indicates that there is correlation and geographic linkage between cyberspace communications and the real-world events. However, some cyberspace representation maps or information landscapes may be distorted from reality to degrees that depend on the media communication channels and varies by topics. As a pilot study of mapping cyberspace to real space, this chapter presents two case studies on visualizing information landscape in cyberspace and also addresses some limitations and suggestions for future research in this domain.

Detecting events from the social media through exemplar-enhanced supervised learning

ABSTRACT Understanding and detecting the intended meaning in social media is challenging because social media messages contain varieties of noise and chaos that are irrelevant to the themes of interests. For example, conventional supervised classification approaches would produce inconsistent solutions to detecting and clarifying whether any given Twitter message is really about a wildfire event. Consequently, a renovated workflow was designed and implemented. The workflow consists of four sequential procedures: (1) Apply the latent semantic analysis and cosine similarity calculation to examine the similarity between Twitter messages; (2) Apply Affinity Propagation to identify exemplars of Twitter messages; (3) Apply the cosine similarity calculation again to automatically match the exemplars to known training results, and (4) Apply accumulative exemplars to classify Twitter messages using a support vector machine approach. The overall correction ratio was over 90% when a series of ongoing and historical wildfire events were examined.

Toward Online Linguistic Surveillance of Threatening Messages

Threats are by definition a communicative act, and as such, it becomes an essential obligation of the field of communication to ascertain what they communicate, and when they communicate imminent credible risk. This paper proposes a research- and theory-based set of over 20 potential linguistic risk indicators through which corpora of actual online threat messages may discriminate credible from non-credible threats. Some of these are obviously easier to code automatically than others, but all can be rated, and used as potential construct validation criteria. Rating scales are proposed, existing threat corpora are identified, and some prospective computational linguistic procedures are identified. Implications for ongoing threat surveillance and its applications are explored.

Sparsity and normalization in word similarity systems

We investigate the problem of improving performance in distributional word similarity systems trained on sparse data, focusing on a family of similarity functions we call Dice family functions (Dice 1945), including the similarity function introduced in Lin (1998), and Curran (2004), as well as a generalized version of Dice Coefficient used in data mining applications (Strehl 2000:55). We propose a generalization of the Dice-family functions which uses a weight parameter α to make the similarity functions asymmetric. We show that this generalized family of functions (α systems) all belong to the class of asymmetric models first proposed in Tversky 1977, and in a multi-task evaluation of 10 word similarity systems, we show that α systems have the best performance across word ranks. In particular, we show that α-parameterization substantially improves the correlations of all Dice-family functions with human judgements on three words sets, in2 Gawron and Stephens cluding the Miller-Charles/Rubenstein Goodenough word set (Miller and Charles 1991, Rubenstein and Goodenough 1965).