Using text mining to quantify qualitative data in literature review: A feasibility study

Background/Objectives:

Existing literature review of qualitative studies is usually conducted manually and subjectively. The results are not quantified for an easier interpretation. We aim to develop an efficient approach by employing text mining techniques to provide quantitative evidence for  the literature review of qualitative studies.

Methods:

Text mining technology and sentiment (emotion) analysis were used. We retrieved 15 qualitative studies published in English prior to 2011 on HIV-positive  parents’ disclosing  their infection to their children, from which we extracted and parsed 600 quotations. A dictionary was developed by combining the terms in public sentiment dictionary and the terms identified by researchers in the field of HIV disclosure. and then used to measure the emotion status (positive/negative) of each quotation. We also developed techniques of identifying and visualizing the most common terms in the quotations and the hierarchical relations among them. Finally we compared the results provided by text mining approach and traditional approach.

Results:

The score of emotion status of each quotation ranged between -5 to 5 with a mean of 0.7, suggesting that children’s emotional responses to parental HIV disclosure were not totally negative. This result was consistent with previous literature review. The top three terms were “Love”, “Kind”, and “Care”. Latent association analysis between terms shows that the terms “Sick ”, “Sad” and “Cry” appeared together in expressing negative emotions, while “Glad”, “Okay” and “Good” bonded in expressing positive emotions.

Discussions:

The current study demonstrates the feasibility of applying text mining in literature review. Given the rapid growing numbers of published empirical studies and tremendous qualitative data (texts) in social media, an automatic or semi-automatic literature review approach has the potential to save remarkable human resources and provide valid, reliable and quantified results that can be widely used in public health research.