Sentiment Analysis for Interview Transcripts

Introduction

The most common way for researchers to get into depth and obtain detailed insights is through one-on-one interviews and the analysis of their transcripts. Interviews can provide a wealth of information. 

Transcripts of such interviews aid in qualitative analysis. However, such analyses are subjective and rely heavily on the authors’ or researchers’ points of view. 

Thus it becomes necessary to analyze these transcripts in a way that will provide objectivity to the interpretation of personal interviews and derive insights.

 A known technique of Sentiment Analysis from the domain of Natural Language Processing with the help of which emotion of a text can be detected.

Roadmap for Sentiment Analysis on Interview transcripts work

To keep us on track we should lay out a brief guide. The following steps form the basis for any sentiment analysis once we have a problem and model in mind:

  1. Determine the aim of the analysis
  2. Conduct the Interviews (If required)
  3. Prepare the interview transcripts
  4. Decide the aspects/set of questionnaires according to which transcript’s content is to be segregated.
  5. Create the dataset
  6. Data Preprocessing
  7. Build Sentiment Analysis
  8. Analyze the results

A dataset can be curated by segregating the transcripts based on a set of questionnaires or based on several aspects of the topic the interview is related to. So we need several interview transcripts. One sample interview transcript can be found here. We classify the transcripts considering the questions to us:

Q1

Can you please tell me a bit about yourself?

Q2

Can you please tell me a bit more about what attracts you to working for XYZ Ltd?

Q3

Can you please outline an issue you’ve had to deal with in the past and how you tackled it?

Q4

Could you talk me through a real situation where you have shown the ability to prioritize and meet deadlines?

Q5

Describe yourself in three words, why?

Q6

What would you say is your greatest weakness?

Q7

Where do you see yourself in five years’ time?

Similar transcripts for 10 different people are taken and a dataset can be curated. (Link to the dataset) If any answer is missing, it is considered to be a missing value. Any missing value present is not imputed to maintain the originality of the transcripts.  When the dataset is ready, sentiment analysis is applied to the answers and the sentiment values are averaged to get the overall sentiment of the interview. 

Different models and techniques can be used for the analysis, one of them being VADER. Valence Aware Dictionary for Sentiment Reasoning is sensitive to both polarity (positive/negative) and intensity or strength of emotion. Sentiment Analysis using VADER relies on a dictionary that maps words or vocabulary of a language features to emotion intensities known as sentiment scores. The total sentiment score of a text can be obtained by summing up the intensity of each sentence in the text. VADER takes in a string as input and returns a dictionary of scores in each of four categories:

  •       negative
  •       neutral
  •       positive
  •       compound (computed by normalizing the above 3 scores)

The compound score ranges from -1 to +1. A score closer to +1 indicates strong positive sentiment whereas a score closer to -1 indicates strong negative sentiment. A value closer to 0 is an indication of neutral sentiment.

Once the sentiment scores are obtained, exploratory data analysis or several machine learning models can be run on the sentiment scores to find patterns and correlations among the interviews. One way of analyzing the result is averaging the sentiment score for each aspect/question to get an understanding of which aspect/question plays a vital role in the interview. A word cloud as shown below can also be plotted to find the frequency and importance of words in the text. This helps to extract features that influence the sentiments of the interview and provide a statistical base to it.

Q. No.

Interview/ Transcript Question

Compound Score

Q1

Can you please tell me a bit about yourself?

0.35

Q2

Can you please tell me a bit more about what attracts you to working for XYZ Ltd?

0.71

Q3

Can you please outline an issue you’ve had to deal with in the past and how you tackled it?

0.18

Q4

Could you talk me through a real situation where you have shown the ability to prioritize and meet deadlines?

0.16

Q5

Describe yourself in three words, why?

0.52

Q6

What would you say is your greatest weakness?

0.33

Q7

Where do you see yourself in five years’ time?

0.56

The dataset, code, and results can be found here. 

References:

  1. https://towardsdatascience.com/sentimental-analysis-using-vader-a3415fef7664
  2. Parmar, Manojkumar & Maturi, Bhanurekha & Dutt, Jhuma & Phate, Hrushikesh. (2018). Sentiment Analysis on Interview Transcripts: An application of NLP for Quantitative Analysis. 10.31235/osf.io/qm8et. 
 Save as PDF

Leave a Comment

Your email address will not be published. Required fields are marked *