X-Message-Number: 24033 Date: Mon, 03 May 2004 02:52:19 -0400 From: "Kevin Q. Brown" <> Subject: CryoNet Ratings and Digest Filtering Update My thanks to everyone who has taken the time to provide ratings for CryoNet messages. The CryoNet code now uses those ratings to compute reputation scores for each recent poster. People who consistently post messages that are regarded poorly by other CryoNet readers now will find that their messages are excised from the daily digests, while still remaining accessible in the archives. One of my next tasks is to explain in detail how those computations are made, which may be as complex as coding them! Upon completing that documentation I'll add a pointer to it from one of the rating-related web pages rather than posting a long, detailed, mind-numbing treatise to CryoNet. For the curious, though, here is an overview. Briefly, the code attempts to compute the likelihood of spam from a particular poster by calculating the typical percentage of ratings in each category ("Offtopic", "Interesting", etc.) for his/her postings, multiplying those values by the "spaminess" weight for each category, and adding the results. The initial "spaminess" weights are: Flamebait => 0.8, Boring => 0.6, Informative => 0.1, Interesting => 0.1, Insightful => 0.1, Redundant => 0.65, Funny => 0.4, Normal => 0.3, Offtopic => 0.9, Poor_Subject_Line => 0.4 Thus, a person whose posts always rate 100% "Informative" will have a probability of spam of 0.1 whereas someone whose posts always rate 100% "Flambait" will have a probability of spam of 0.8. The reputation is just 1 - the probability of spam. The "Informative" poster thus has a reputation of 0.9 whereas the "Flamebait" poster has a reputation of 0.2. A poster with a reputation less than one third currently is considered a spammer. Admittedly, those "spaminess" weights are somewhat arbitrary. Eventually they will be fine-tuned by applying Bayes Rule to the computations. Meanwhile, if you are curious about the reputation score the system currently computes for your postings, try URL: http://www.cryonet.org/cgi-bin/reputation.cgi This requires your "rater code" (either entered in a web form or retrieved from a cookie), so only you can see your current score. FYI: Currently, no reputation can be below 0.1 or above 0.9 because those are the minimum and maximum "spaminess" weights. Bear in mind that the reputation scores are recomputed only once each day, whereas CryoNet subscribers can update the message ratings throughout the day. If your reputation score looks way out-of-line with the ratings of your postings, though, please let me know. Kevin Q. Brown FYI: The "Rate This Digest" hyperlink, which immediately follows the list of all messages in each daily digest, provides a convenient way to both view the digest's messages and rate them. Alas, I haven't (yet?) implemented the ability for a poster to rate his/her own messages, as suggested in message # 23987. Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=24033