Please enable JavaScript.
Coggle requires JavaScript to display documents.
SLR Problems/Challenges (Select Relevant Studies (Text Mining/Visual Text…
SLR Problems/Challenges
Time
Rapid Reviews (RRs)
Developing Protocol
Authors must make important SLR planning decisions
Review research questions
Review selection criteria
Review search strategy
Define study selection
Extract data
Define Quality assessment
Choose data synthesis method
Visual Texting Mining
Identifying the most appropriate research questions
SLR Planning Based on Suffix Tree Clustering (SLRP-STC) is a VTM strategy designed to identify common phrases in a collection of studies and use these phrases as the basis for creating clusters
Searching for Evidence
Inadequate search
strategy
Heterogeneity of SE terminology
Limited range of indexing terms
Two types of automation
(1) automation to help in search string generation
VTM to recommend string terms
AI to creation and calibration of search strings
Hill Climbing (HC) algorithm (it just uses IEEEXplore as source)
Machine Learning
It uses statistical inference to learn which studies are relevant
The algorithm generates the search string using a data-driven approach based on terms from title, abstract and keyword of the currently accepted papers
Then, the algorithm performs snowballing using Scopus
Unified Search Tool for DLs (ACM, IEEE, Scopus, WoS, Science Direct, Scopus and Google Scholar). But due to the limited access to the full text of studies, the tool only retrieves title, keywords, and abstract (Feng et al.)
(2) automation to help during the search execution
Select Relevant Studies
Large number of
studies to be analysed
Text Mining/Visual Text Mining
TM helps filter relevant studies during the first stage of selection (reading abstracts), thereby reducing the set of studies that researchers must examine examined in the final selection (full-text).
VTMto generate visualizations of the primary studies to support the selection activity (document organisation in maps and clusters)
Sampling
Using bibliometric approximation
Machine Learning
FASTREAD combines and parameterizes the most efficient active learning algorithms to support study selection when there are a large number of candidate studies
Expert-Driven Automatic Methodology (EDAM)
It generates an ontology using candidate studies and then classifies studies using that ontology, allowed researchers to spend their effort on analysis and discussion rather than on classification
Concept Maps
To summarize a complex structure of textual information, help researchers identify the most relevant studies
It can be supported with Natural Language Processing to reduce the effort required
Rank studies in decreasing order of importance for an SLR with respect to the terms in the search string
Extracting Data
Inconsistent data formats and
paper designs
Extractive Text Summarization
partitions a document into a set of topics and then choose the most relevant sentences for each topic (TextTiling)
Torres et al. Tool has as main objective to locate sentences that specifically represent study results
Regular Expressions
Text Mining and Machine Learning
TM and machine learning identify section headings (paper structure) from research studies. This approach uses a statistical analysis of the most frequent word- s/phrases in the section headings to build the structure, which serves for subsequent automatic extraction of data
Synthesizing the evidence
Papers may not provide
all the necessary information
Differences between studies is a
source of heterogeneity
Visual Text Mining
To support the categorization and classification of studies
Association rules
A strategy for extracting multiple patterns using association rules analysis. Its goal is to generate relationships, associations, correla- tions or frequent patterns between the attributes of a collection of studies