Drumbeat/MoJo/hackfest/berlin/projects/followthis
From MozillaWiki
Contents
Project Name: FollowThis
Project Lead(s): Matt Terenzio
Big Goal for MoJo Hackfest:
- Ship some usable code. (achieved)
- Learn how to manage an Open Source project.
- Work with others on related projects. (achieved)
- Drink heavily. (achieved)
Key steps toward goal:
- 1. need to be able to extract RDFa, Microformats from pages. (working)
- 2. Need to be able to use NLP to extract entities if Semantic metadata is not present. (adopt and contribute to metameta project for this)
- 3. Need to be able to store and query the metadata. (am currently able to query the RDF triplestore but need to hone queries)
- 4. Need a solid UI for users to be able to interact with the service. (getting there)
- 5. A crawler for the news sources would be nice. (deferred to version .2)
Pending needs:
- Important:Need to make a button that is an embeddable widget for ease of deployment
- I have a working bookmarklet but it needs work. JQuery help. (still need a session with jquery expert)
- Totally clueless on entity extraction from pages that don't have semantic metadata. (solved somewhat)
- Also need to figure out SPARQL and the best persistent data store for RDF. (Laurian gave me some good starting points)
Link for more info:
- rNews has been brought up. I've installed the RDFa distiller from W3C and you can use it to distill RDFa from pages.
Example which distills a page with rNews in it: To use it, just call: http://followth.is/cgi-bin/RDFa.py?uri=uri-of-we-page-youwant-to-distill
- (update: Matt has a better entity extractor than this using Stanford NLP -- will use that) To extract keywords from some text I set up a CGI script that does so if you feed it text.
It should accept posts to that URL as well as gets.
- First pass at a readability-like way to extract the article text and headline from a web page:
- Another endpoint that distills RDFa froma web page (this one in PHP)
- A SPARQL endpoint for the triplestore of rNews data
http://followth.is/transform/sparql/
Link for demo:
Link to source code:
Where from here:
- Though code is in working form, it is necessary to clean and organize a few parts for better forward maintainability and extension
- Continue to work on open alternatives to some of the portions that use third party APIs
- Documentation for both developers and users
- Promote rNews adoption
- Deploy to at least one news site by end of 2011
Project Status
- Though code is in working form, it is necessary to clean and organize a few parts for better forward maintainability and extension
- Code still relies on third party APIs. Need to offer open source alternative to those parts
Collaborators
The following folks helped with this project:
- Laurian/ How to model data for RDF storage and how to query that data using SPARQL
- Raynor/ TF-IDF (term frequency–inverse document frequency)
- Laurian/Raynor/ Cosine Similarity concepts for comparing documents
- Matt (BBC)/ Entity extraction with Stanford NLP
Also really needed these perspectives:
- Jordan/ What constitutes a valuable difference between documents from a user or journalist perspective
- Chris/ perspective of CMS usage from an editorial standpoint
- Mark/ Open Source project strategy
Next steps
- From here I would like to:
- Continue to work on open alternatives to some of the portions that use third party APIs, including actively contributing to metameta project which is solving some of same
- Documentation for both developers and users
- Promote rNews adoption
- Deploy to at least one news site by end of 2011
Places where this project might be tested include:
- Since I'm still at TheHour.com, I plan on testing it there VERY soon.
- Zeit, Guardian, Boston Globe . . . any news site with text articles is really appropriate