class: center, middle, inverse, title-slide # NLP Candidate ## Ryan Wesslen ### UNC Charlotte ### Sept 17, 2018 --- # Agenda - About Me: 5 min - My Research: 10 min - Topic Modeling to Analyze CFPB Complaints: 30 min - Discussion & Questions: 15 min --- #### UNC Charlotte PhD Candidate - Computing & Information Systems (Computer Science) - UNC Visualization Center, Pacific Northwest National Laboratory, Data Science Initiative, Project Mosaic #### Prior Education - UNC-Chapel Hill (BA in Econ, 2003), NYU (MA in Econ, 2011), NCSU (MS in Fin Math, 2009) #### BofA (2009-2014) / Hawkeye (2014-2015) - Credit Risk and Marketing Analytics and Strategy - GRMAP, Small Business Credit Risk, Auto/DFS Scorecard Modeling #### Teaching & R/R Studio enthusiast - Taught UNCC workshops ([github](https://github.com/wesslen)) in R for text, social media, data viz. - Tentatively teaching master's Visual Analytics course for UNCC DSI in Spring 2019 --- class: center, middle # My Research <img src="./img/research.png" width=700 height=500> --- ## Event Detection in Social Media <div align="center"> <img src="./img/crystalball.gif" width=700 height=450> </div> [CrystalBall](https://www.researchgate.net/profile/Isaac_Cho/publication/324598122_CrystalBall_A_Visual_Analytic_System_for_Future_Event_Discovery_and_Analysis_from_Social_Media_Data/links/5ad7bd1f458515c60f588bb8/CrystalBall-A-Visual-Analytic-System-for-Future-Event-Discovery-and-Analysis-from-Social-Media-Data.pdf) Paper --- ## Misinformation using Visual Analytics <div align="center"> <video width="600" height="450" controls> <source src="./img/verifi.mp4" type="video/mp4"> </video> </div> [ICWSM](https://aaai.org/ocs/index.php/ICWSM/ICWSM18/paper/view/17853) Paper --- ## Explainable AI / Interpretable ML <div align="center"> <img src="./img/xai.png" width=700 height=400> </div> Source: [DARPA XAI](https://www.darpa.mil/program/explainable-artificial-intelligence) --- ## Visual Interactive Labeling (CHISSL) <div align="center"> <video width="600" height="450" controls> <source src="./img/chissl.mp4" type="video/mp4"> </video> </div> Paper under revision --- ## Applied Text Analyses .pull-left[ <div align="center"> <img src="./img/tmm.png" width=500 height=300> </div> For more technical, see ["Computer Assisted Text Analysis"](https://arxiv.org/abs/1803.11045) paper ] .pull-right[ - **Management**: Corporate websites on recruitment content (["MNE"]() paper) - **Organizational Science**: Open-ended survey on manager leadership (["TMM" paper](https://link.springer.com/article/10.1007/s10869-017-9528-3), left) - **Social Psychology**: Twitter profiles (["Bumper sticker"](https://aaai.org/ocs/index.php/ICWSM/ICWSM18/paper/view/17834) paper) - **Communications**: Charlotte Protests tweets (next slide) - I also have text experience with research absracts, patents, news articles, emails, ... ] --- <div align="center"> <img src="./img/charlotteprotests.png" width=800 height=550> </div> [IC2S2 2017](https://wesslen.github.io/assets/documents/presentations/IC2S2-HotIssue-Charlotte.pdf) Charlotte Protest Presentation --- class: center, middle # Text Analysis Methods
--- class: middle  Source: Pablo Barbera --- class: center, middle <div align="center"> <img src="./img/cfpb.png" width=500 height=450> </div> ## CFPB Narrative Complaint Competitive Analysis --- .pull-left[ Analyze 46,590 CFPB Complaints with *narratives* between **March 2015** to **August 2018** for BAC, JPM, WFC, and CITI. <div align="center"> <img src="./img/total-counts.png" width=360 height=250> </div> Use **topic modeling** to identify prevalent themes via unsupervised approach. Exploratory, document summarization technique needing *little assumed knowledge*. ] .pull-right[ <img src="./img/dendrogram.png" width=270 height=600> ] --- class: center, middle  --- # Example: Topic 18 <div align="center"> <img src="./img/example-topic.png" width=532 height=488> </div> --- # Topic 18: "Bonus, offer, points, promotion"
--- class: center, middle  --- class: center, middle  [Structural topic modeling]9http://www.structuraltopicmodel.com/) --- class: center, middle  --- class: center, middle # Questions & Discussion: Thanks! [github.com/wesslen](https://github.com/wesslen) [wesslen.github.io](https://wesslen.github.io) --- class: center, middle 