Hi all, in the past few weeks I was experimenting and trying to classify the spots comment using AI. The workflow is the following: Spot is received and forwarded to the AI engine The AI engine returns an "offensive" score probability and category. The score is analyzed and an algorithm decides if it is the case to block the source ip of the spotter. The decision is taken considering several factors (scores of the single spot, score of previous spots from the same ip, rate, ect). If the algorithm concludes that the ip is to be blocked it inserts it in a text list. The bad ip is removed from the list after some time. Since not all the spots in the PC61/PC11 packet contain the source ip, when the source ip is not available, the same logic is applied to the source node. In my cluster I did a further step by automatically inserting/removing the AI detected bad ip and bad node into the respective bad list to drop these spots. I did some tests during the last weekend while the bad spots from the usual individual were coming and the model works pretty well. It took less than 5 minutes to block bad ip and bad nodes. The interesting part is that AI is able also to determine the context: even if in the spot there are no bad words, the comment could result offensive, like (real) example below: +-----------------+---------------------+----------------------------------+-------------+----------+---------+-----------+-------------+---------+ | ip | Data | comment | source_node | dx | spotter | frequency | IA_detected | Blocked | +-----------------+---------------------+----------------------------------+-------------+----------+---------+-----------+-------------+---------+ | 188.217.216.80 | 2026-01-13 19:22:27 | real donkey no brain | S50CLX | PG4T | IC8TEM | 21074 | Y | Y | | 85.148.197.200 | 2026-01-13 19:47:31 | look at yourself monkey | WA9PIE-2 | IC8TEM | PG4T | 21074 | Y | Y | Please note that the AI engine is not "in-line" with the cluster. This means that the offending spot is still visible on the cluster, but once detected, subsequent spots from the same ip (or node, but see below, [NOTE_1]) are blocked. The AI is not always exact at 100%. Consider also the meaning of different terms in different languages. I.e. "ARMI" in italian means "weapons" but also "Italian sailors' amateur radio association", so it's not easy to understand the context of a spot like "ARMI district" ;-) For this reason I have foreseen a white-list in the algorithm. I'm publishing the badip.ai and badnode.ai files here: http://badlist_ai.iz2lsc.eu:4000/badlist The files are updated every 2 minutes. If you want to try it I suggest you to insert in dxspider crontab the download of the badip.ai file every 5 minutes and reload the badip list. A sample crontab is ######### download bad ip AI list every 5 min ############################################### 0,5,10,15,20,25,30,35,40,45,50,55 * * * * spawn('cd /spider/local_data; wget -qN http://badlist_ai.iz2lsc.eu:4000/badlist/badip.ai') 1,6,11,16,21,26,31,36,41,45,51,56 * * * * run_cmd('load/badip') ############################################################################################## [NOTE_1]: Unfortunately at the moment DXspider does not support dynamic addition/removal for badnodes in the same way as badip, so I don't know how to automatically apply the badnode file to the dxspider * Maybe Dirk or Kin can help here.* In any case if you also want to download the badnode.ai file, this is the crontab, but then you have to apply it manually with set/badnode or unset/badnode for every node in the list. ######### download bad node AI list every 5 min ############################################### 2,7,12,17,22,27,32,37,42,47,52,57 * * * * spawn('cd /spider/local_data; wget -qN http://badlist_ai.iz2lsc.eu:4000/badlist/badnode.ai') ############################################################################################## Please note that this is still in the *"test phase"* so please report to me any issue or problem. I hope this could help a bit with preventing the shame we observed in the last week on the cluster. 73 Andrea IZ2LSC -->