Changes between Version 38 and Version 39 of SimilarityFunction

Timestamp:

06/10/09 16:03:30 (5 months ago)

Author:

niels@… (IP: 130.161.159.240)

Comment:

--

SimilarityFunction

@@ -125 +125 @@
 
 == Dataset ==
-Parsing superpeerlogs in order to get insight into availible data.
+First started by parsing superpeer logs in order to get insight into availible data (See SuperpeerLogs). 
 
-Stats extracted from logs:
-[[Image(tribler_daily_usage.png)]]
-The usage of tribler spikes in 2007, after slashdot and bbc news published articles on it.
+Then a subset was created using the top-50000 users with the most downloaded files.
+This subset has 252.469 items and 50.000 users. Using tf/idf 31.906 items were assigned to a category. This helps evaluating the performance of more complex similarity functions and was done manually. First using the tf/idf the more frequent terms were discovered. Which were used to create a list of categories. All items matching all or a combination of the terms of a category where written to a file. These files were then checked manually and incorrect items were removed/disabled.
 
-[[Image(tribler_first_last.png)]]
-This graph is generated using the first and lastseen date of each permid.
-
-[[Image(tribler_top50_cumul.png)]]
-A strange graph, the tribler top 50 most active downloaders consists of 2 groups. The first group is a very old one and used the program in 2006. The new group are users starting in 2009 and are using the 4.5 client. But no active users in between?
-
-[[Image(tribler_top50_hour.png)]]
-Used to detect bots in the top50 downloaders. But possibly due to resolution of the superpeerlogs and the buddycast protocol (send message every 4 hours) this graph says nothing.