OpenSocial Gadget Contest

Adding new features to Profiles and/or VIVO

Using Twitter data in combination with search results

Proposal Status: 

This gadget is designed to use the Twitter API and RDF data from search results conducted on the UCSF Profiles site to return relevant data about search. Using the query terms, the gadget will pull relevant data from currently trending tweets and display these results along with the search results. We would like to be able to highlight the differences in tweet content so that tweets featuring links to articles, resources, and conferences are given priority. 

Comments

This wouldn't be difficult to build. For me, the question is how to filter tweets so they're the most useful.

For example, I searched for "lung cancer" on Twitter, and here's the top 5 results:

Are these the kind of results you were thinking of?

I think it would be interesting to 'highlight' tweets that are relevent to the logged in user.  This could be done by creating a relvancy score between any external content (a tweet) and a person (the logged in user) based on the persons RDF.  

For example, if a tweet comes by that mentions 'stroke' and the logged in user has 'stroke' listed as a vivo:freetextKeyword, or in their vivo:overview, then that tweet would be highlighted.

Ideally we would create a general purpose relevance score that would show how relevant some arbitrary content is to a person.  I can see how this could be done via some search engine components and a weighting being applied to properties of the vivo ontology.

I think that this is an interesting idea using Twitter as a resource for research-related content (e.g., articles, resources, and conferences).

I agree with Eric and Anirvan that it would be useful to filter the tweets in a way that matches the content with user interest such as research topics (UCSF Profiles keywords) or specific disease areas. 

I'd like to learn a bit more about how trending topics would be used here. It is often hard to understand what they are about. However, a high percentage of trending topics are hashtags, so it might be helpful to leverage healthcare and disease hashtags for this project idea (http://www.symplur.com/healthcare-hashtags/regular/; http://www.symplur...).

The following research paper might also be interesting: "Twitter Trending Topic Classification" (Link http://cucis.ece.northwestern.edu/publications/pdf/LeePal11.pdf) The authors looked at Twitter trending topics and categorized them into general categories (e.g., sports, politics, technology). They conclude that a network-based classifier performed significantly better than text-based classifier on their dataset. 

 

 

 

 

 

 

Selected comments from Reviewer(s):

Reviewer 1: "More relevant to desk researchers, I suspect. Unmoderated data means the relevance of the tweets would be up to the reader to determine."
Reviewer 2: "big effort to do it well I think.  Also, one thing to think about is where do inserts/gadgets make sense from a single investigator viewpoint, or from a thematic viewpoint.  This may be more relevant for those exploring the concept search in profiles 1.0 where the twitter content maps onto the stream of recent pubs on a topic from UCSF..."

On behalf of Clinical and Translational Science Institute at UCSF, thank you for participating in this contest.

Commenting is closed.