Honors Program

 

Date of this Version

5-2020

Document Type

Thesis

Citation

Leising, C., Nguyen, N., Rhoadarmer, B., Shaffer, K., & Zatorski, G. Natural Language Processing. Undergraduate Honors Thesis. University of Nebraska-Lincoln. 2020

Comments

Copyright Grace Zatorski 2020.

Abstract

Currently, the entire process for a single interview takes between 10 and 15 hours to complete. The team developed a system that can take in an audio file, automatically transcribe it, and output the transcript in either a markdown file or a word document in 20 to 30 minutes. This will allow transcribers to focus on the words/phrases that are difficult for AWS Transcribe to understand while reducing transcription time by 800-1,800%.

The system also accepts transcriptions in the form of a spreadsheet as input. Once given a transcription to evaluate, the system will remove extraneous words and punctuation before passing it to the analysis and evaluation tools: word analysis and scoring. Interviews are scored by leveraging supervised learning techniques to create models for each particular question. All of the individual scores are aggregated to a value that is 77.45% accurate compared to the analyst's scores. Another analysis tool takes a list of words and phrases and calculates the frequency of the usage of each time and the location of the item in the interview. This tool will allow Talent+ to evaluate the correlation between a variety of patterns and interview scores that could be used to predict success. The system takes about 2 minutes to complete a single interview, an increase of 12,000-18,000% from the manual evaluation method.

Share

COinS