Abstract: Using Text Mining to Find Meaning in Father Narratives (Society for Social Work and Research 25th Annual Conference - Social Work Science for Social Change)

All live presentations are in Eastern time zone.

Using Text Mining to Find Meaning in Father Narratives

Friday, January 22, 2021
* noted as presenting author
Jeffrey Shears, PhD, Professor; Director, Joint Master of Social Work Program, North Carolina Agricultural and Technical State University

Background: This study explores the EHS Father Study qualitative data focusing on the open-ended question, “Talk about your experience with your father as a child.” This was an embedded question in the n=575 EHS Fatherhood Study interview of fathers when their children were 24 and 36-months. This question was part of an open ended interview exploring fatherhood in low-income families. This sample includes biological residential fathers (66%), residential father figures (15%), non-residential biological fathers (15%) and non-residential father figures (4%).

Methods: Using qualitative data, we incorporated a text mining approach to explore fathers’ responses quantitatively. Text mining is a subfield of data mining and is used to extract information and patterns from large sets of textual data which is used to analyze call center data. This method enables us to quantitatively analyze textual data which has traditionally remained in the realm of qualitative analysis approaches. To supplement the process, we used sentiment analysis and topic modeling which provide a computational way to identify and categorize the meaning of textual data. Sentiment analysis is a useful tool to analyze understanding interviewees’ sentiment in the interview responses while topic modeling is a type of unsupervised document classification technique which helps identify hidden topics across many individual interview responses.

Results: This analysis allowed us to use the sentiment score to explore if there were words that fathers used in describing their experiences with their father that were preceded by a negative word. The results indicate that the Bing sentiment scores are closely centered at zero for both the mean and median while the median of the Afinn score is 2. The distribution of the number of words used in individuals is positively skewed, which is evident by the smaller median value of 45 compared to the mean of 73.8. Our hypothesis was that the relationship between the number of words and the sentiment score, that is, the interview participants with a negative sentiment score tend to talk more. The coefficient of -0.0038 (P-value =0.0036) for the number of words on the Bing sentiment score analysis is statistically significant, which indicates that when a father used one additional word, his sentiment scores decreased by 0.0038 on average. From this observation, it can be inferred that fathers who spoke longer felt more negatively about their father. The topic modeling identified several salient topics which include outdoor family relationship, family activities, outdoor activities, teaching and learning.

Implications: This study uses text mining, an innovative analysis approach to examine the power of words and experiential meaning in describing fatherhood in the context of being fathered. The text mining analyses offer initial inquiry to the question of how fathers reflected on being fathered. Intergenerational parenting suggests that how men were fathered may predict future parenting behaviors in men. Future research will allow us to use these methods to quantitatively analyze how men speak about their fathers and quantitative measures of fathering and childhood outcomes of children.