Current Research Topics, March 1996


Introduction to filtering


	Streams of Electronic Messages: News, Email

	- Mandatory (must read, internal priority)

	- Educational (should read, internal interestingness)

	- Entertainment (read when possible, internal interestingness)


	- NOISE: thread mutation, flame, short comments



	No time to read it all.

	Classification and selection takes time.

	Short messages. Must read most of it to know its classification.



	HUMAN FILTERING

	cognitive	Subject:	keywords, phrases, idioms

	social		Sender:		commitment, quality, content

	social		To: Cc:		personal, group, mailing-list

	economic	Length:		content, reding time, transfer time

	
	Combinations:	Sender & Length -> contents

			Subject & Length -> contents (bird.jpg, 122 bytes)


	Email is classified twice: 
		
		when received for priority, 

		when stored for retrieval


	FK's matrix

			Work		Private

	Personal	1		2

	Group		3		4


	Subdivisions: orders, questions, requests, information, advertising.
	
		Orders, Information and Advertising mostly statements.

		Questions and Requests mostly questions.


	IDIOMS support string matching

		Re:	followup marker (references field)
		WB	wants to buy
		FS	for sale
		RFI	Request For Input
		CFP	Call for papers

	Subject abuse:	======**** IMPORTANT ****======


			MESSAGE FLOW		INFORMATION RETRIEVAL
	purpose		serendipity		search
	query		good-?-bad		information
	source		upstream		downstream (repository)
	novelty		recent			archived
	noise		high noise		low noise
	structure	threads			articles

		       Alta Vista Search Engine


	Complexity of human classification

	- knowledge, stress, interests, priorities

	


A survey of filtering systems

	SLIDES


	GroupLens	MIT, Univ. of Minnesota



The PEFNA system