You can download the PDF version here.

Research Interests

Natural Language Generation, Text Summarization, Truthfulness and Factuality

Education

Bachelor of Science in Information Science, Statistics
Aug 2017 - Dec 2020
Cornell University
Magna Cum Laude

Work Experience

Data Scientist, Bank of America, Erica Conversational AI Research & Development
Plano, TX, July 2021 - Present
Created Semantic Role Labeling system specific to Conversational AI to improve the contextual understanding of chatbots
Improved generative dialogue summarization model for call centers by creating a summarization dataset focused specifically on task-driven dialogue summaries for customer service and invented a hybrid extractive-abstractive modeling technique for real-time summarization
Improved machine-translation system using weakly supervised methods of data generation for Spanish for Erica solution

Global Technology Summer Analyst, Bank of America
Remote, May 2020 - Aug 2020
Team lead creating a forecasting model for ATM utilization during the pandemic, reduced MAE from 8.6% to 6.3%

Data Analyst Intern, Corning Incorporated
Corning, NY, May 2019 - Aug 2019
Improved emerging trend identification by analyzing news data using topic modeling to track the rise and fall of industry trends

Current Research Projects

Abstractive Dialogue Summarization
Creating an Issue-Resolution Summarization system for call center calls with the issue a customer is calling about and how the agent resolved a call
Built a Bart-based summarizer fine-tuned on DialogSum & XSum datasets
Employed methods to generate more faithful/truthful summaries such as training with a contrastive loss function and re-ranking beams by automatic faithfulness metrics

Title Generation
Research methods of extreme summarization to provide short descriptions for the purpose of extracting the main topic of a text
Created topic summarization system using the following methodology Used fine-tuned issue-resolution Bart to generate 3 summary beams for about 15k call transcripts
Used Few-shot label generation with MPT-7B based off of Unlabeled Data Generation methodology to generate extreme summaries from the issue-resolution summaries
Used transfer learning to train Bart to learn extreme summaries from original dialogue text.

Semantic Role Labeling
Researching semantic role systems for dialogue systems to improve contextual understanding of low-resource systems
Proposed new semantic role schema specifically for chatbots
Demonstrated improved contextual understanding of chatbot when using the semantic role system to gain a better underlying understanding of language

Machine Translation
Creating Spanish version of Erica by using automatic translation to create English version of utterances. Established novel financial Spanish-English translation dataset and nstituted a weak supervision loop to improve quality and quantity of training data

Teaching

Teaching Assistant, Introduction to Data Science (INFO/CS 2950), Spring 2020 & Fall 2020
Teaching Assistant, Introduction to Computing Using Python (CS 1100), Spring 2019 & Fall 2019

Membership

Women in Computing at Cornell (2017-2020)
Information Science Student Association (2018-2020)
Women in Data Science at Bank of America (WiDS) (2021 - present)

Leadership/Service

Executive Board Member, Women in Data Science at Bank of America (2022 - present)
Program Lead, Girls Who Code of North Texas Summer Immersion Program (2023 - present)
Mentor, The Coding School (2021 - 2022)

Patents

  1. “Selection System for contextual prediction processing versus classical prediction processing”. US Patent Application No. 17/993,048, filed November 23, 2022.
  2. “Action-topic Ontology”. US Patent Application No. 17/993,038, filed November 23, 2022.
  3. “Semantic frame builder”. US Patent Application No. 17/993,029, filed November 23, 2022.
  4. “Dynamic semantic role classification”. US Patent Application No. 17/993,019, filed November 23, 2022.
  5. “Dual-pipeline utterance output construct”. US Patent Application No. 17/993,013, filed November 23, 2022.
  6. “Iterative Processing System for Small Amounts of Training Data”. US Patent Application No. 18/199,073, filed May 18, 2023.
  7. “Multilingual Chatbot”. US Patent Application No. 17/993,063, filed November 23, 2022.
  8. “Performance Optimization for Real-time Large Language Speech-to-text Systems”. US Patent Application No. 18/204,981, filed June 2, 2023.
  9. “Call center voice system for use with a real-time complaint identification system”. US Patent Application No. 18/144,925, filed May 9, 2023.

Relevant Coursework

Introduction to Data Science, Natural Language Processing, Machine Learning for Intelligent Systems, Machine Learning for Data Science, Statistical Computing, Data-Driven Web Applications, Interactive Information Visualization

References

David Mimno, Cornell University
Emad Noorzidaeh, Bank of America