An overview of the literature on assistance dogs using text mining and topic analysis

Abstract

It is said that dogs are human's best friend. On occasion, dogs can be raised and trained to provide additional specific benefits to humans suffering from a range of physical or mental conditions, working as assistance dogs. In this article, we employed innovative techniques to review the vast and constantly expanding literature on the subject, which covers a multitude of aspects. The 450 articles obtained through keyword search on Scopus were initially described in terms of year of publication, geographical context and publication destination, and were subsequently analysed through automated text mining to detect the most important words contained within them. Lastly, a generative model of topic analysis (Latent Dirichlet Allocation-LDA) described the content of the collection of documents, dividing it into the appropriate number of topics. The results yielded interesting insights across all domains, demonstrating the potential of automated text mining and topic analysis as a useful tool to support the researchers in dealing with complex and time-consuming subjects' reviews, integrating the work done with traditional reviewing methods.

Keywords: Latent Dirichlet Allocation; assistance dogs; generative model; service dogs; text mining; topic analysis.

Figure 1

Categories of assistance dogs considered in this review.

Figure 2

Flow chart of the screening process of scientific literature concerning assistance dogs.

Figure 3

Distribution of studies published by 5-year intervals (n = 450). The last interval (2019–2023) includes the first 2 months of 2024, until the date of search (8th of February 2024).

Figure 4

Distribution of studies (n = 450) by country. It should be noted that some studies are included in more than one country. The category other (n = 51) includes Austria, Denmark, Germany, Israel, Estonia, Finland, Hungary, Norway, Spain, Brazil, China, Croatia, Mexico, South Korea, Argentina, India, Luxembourg, Morocco, Palau, Singapore, South Africa, Switzerland, Trinidad and Tobago, and Ukraine.

Figure 5

Distribution of studies (n = 450) by subject area. Some studies are included in more than one subject. The category other (n = 16) includes Chemical Engineering, Chemistry, Computer Science, Economics, Econometrics and Finance, Physics and Astronomy, Energy, Immunology, and Microbiology.

Figure 6

Barplot representing the frequency rate of the 25 most relevant (TF_IDF) terms in the whole collection of documents.

Figure 7

Barplot representing the inverse document frequency of the 25 most relevant (TF_IDF) terms in the whole collection of documents.

Figure 8

Barplot representing the list of the 25 most relevant (TF_IDF) terms in the whole collection of documents.

Figure 9

Line plot representing the performances of four different methods for estimating the most accurate number of topics in the corpus of documents.

Figure 10

Topics emerged and categories identified.