Flying to the Top, One Tweet at a Time: Using Social Media to Rank Online Search Results
Robyn B. Reed, MA, MLISCo-authors:
Carrie L. Iwema, PhD, MLS Ansuman Chattopadhyay, PhD
Health Sciences Library SystemUniversity of Pittsburgh
Workshops
Website
Software Licensing
Consultations
Molecular Biology Information Service
Online Bioinformatics Resources Collection (OBRC)
http://www.hsls.pitt.edu/obrc/
Resources displayed by keyword ranking
http://www.hsls.pitt.edu/obrc/
Challenges:Many tools exist and increasing in number
User may retrieve several resources
Common question –How do I know which one(s) to
use?
Goal:Provide up-to-date ratings of most
highly regarded resources in bioinformatics
Objectives:
Using social media, design ranking system of OBRC resources
Determine if social media results reflect opinions of bioinformatics experts
Why use the social media??
• No official rankings of bioinformatics tools
• Opinions of several people
• Social media data has many applications
http://beta.socialguide.com/
Methodology
Wrote 5 research questions
Common bioinformatics queries
Each question listed 3 possible resources to accomplish that task
Resources wereranked using social media data
Experts (2) independently ranked resources
Methodology
Research questions
Methodology – Social Media Ranking
Sources used for data collection
Google Blogs Google Discussions
Google Discussions includes• Forums• Groups• Comments
www.google.com
Twitter considered and removed
• 50% of the resources had zero Tweets• 20% captured non-specific Tweets
Facebook not included
• Concern over private settings
Methodology – Data Sources
Methodology – Social Media Ranking
Searched “all time”
Optimized for most accurate retrieval
• Resource in quotes• Increased specificity, decreased noise• Fewer hits
[(“ucsc genome browser”) AND ( bioinformatics | genome | genetics | genomics | computer | algorithm | software | server | database | computer model | protein | proteomics | proteome | gene | DNA | RNA | sequence | alignment | interactions | structure | modeling | prediction | biochemistry | molecular biology | systems biology | computational biology)]
• Put all OBRC resources in bioinformatics context• Automate the searches
Example of search of UCSC genome browser
Methodology – Search Filter
Results Bioinformatics Tools
Blogs + Discussion Raw
NumbersSocial
Media RankExpert 1 Rank
Expert 2 Rank
CPHmodels 49 2 2 23-D protein prediction ESypred3D 17 3 3 3
SWISS-MODEL 228 1 1 1 IDT SciTools 4 2 2 2
PCR primer design Primer3 728 1 1 1 Primer Design Assistant 0 3 3 3 DIANA-microT 12 1 1 2
microRNA target design miRGator 9 2 2 3
siRNA target finder Ambion 3 3 3 1 ClustalW 1494 1 1 3multiple sequence alignment
ECR Browser 8 3 3 1 Tcoffee 63 2 2 2 Ensembl 3070 1 3 2
genome browsers NCBI Map Viewer 56 3 2 3 UCSC Genome Browser 928 2 1 1
Bioinformatics Tools
Blogs + Discussion Raw
NumbersSocial
Media RankExpert 1 Rank
Expert 2 Rank
CPHmodels 49 2 2 23-D protein prediction ESypred3D 17 3 3 3
SWISS-MODEL 228 1 1 1 IDT SciTools 4 2 2 2
PCR primer design Primer3 728 1 1 1 Primer Design Assistant 0 3 3 3 DIANA-microT 12 1 1 2
microRNA target design miRGator 9 2 2 3
siRNA target finder Ambion 3 3 3 1 ClustalW 1494 1 1 3multiple sequence alignment
ECR Browser 8 3 3 1 Tcoffee 63 2 2 2 Ensembl 3070 1 3 2
genome browsers NCBI Map Viewer 56 3 2 3 UCSC Genome Browser 928 2 1 1
Results
Bioinformatics Tools
Blogs + Discussion Raw
NumbersSocial
Media RankExpert 1 Rank
Expert 2 Rank
CPHmodels 49 2 2 23-D protein prediction ESypred3D 17 3 3 3
SWISS-MODEL 228 1 1 1 IDT SciTools 4 2 2 2
PCR primer design Primer3 728 1 1 1 Primer Design Assistant 0 3 3 3 DIANA-microT 12 1 1 2
microRNA target design miRGator 9 2 2 3
siRNA target finder Ambion 3 3 3 1 ClustalW 1494 1 1 3multiple sequence alignment
ECR Browser 8 3 3 1 Tcoffee 63 2 2 2 Ensembl 3070 1 3 2
genome browsers NCBI Map Viewer 56 3 2 3 UCSC Genome Browser 928 2 1 1
Results
Conclusions:
This system can be used to determine highly regarded tools
Explain that rankings are subjective;
try the top 3-5 resources
Provides patron with a starting point when using the OBRC
Limitations
• Quotation marks can be limiting ifresource >1 word
• Very small part of the total social media
• “Negative” discussion about a resource
Future Directions
• Test > 3 bioinformatics tools/category
• Increase number of expert ratings
• Test applicability of system in areas other than bioinformatics
Special thanks to:
Project collaborators and experts:Ansuman Chattopadhyay, PhDCarrie Iwema, PhD, MLS
Research and academic advisors: Nancy Tannery, MLSRebecca Crowley, MD, MS
Funding from the Pittsburgh Biomedical Informatics Training Program
NLM Grant 3 T15 LM007059-23S1