Content
@
0 reply
0 recast
0 reaction
Nate Maddrey
@nmadd
Does anyone have a good method for filtering out low quality casts from a data set? I'm doing some analysis where I'm pulling a large number of casts and then analyzing engagement and some other stats. But I've found that a lot of replies are basically just spam, so they're throwing off my engagement numbers. Any ideas for how to filter out these spammy casts?
1 reply
0 recast
1 reaction
Victor
@victoreram
Depends on your definition of low quality. A lot of "$DEGEN" tipping to be spam. There are other engagement hacking traits that are subjective like overuse of line breaks (clicking see more -> engagement) or casts that follow a <statement> <line break> <question> structure (which would have many false negatives)
1 reply
0 recast
1 reaction
Nate Maddrey
@nmadd
yeah good point, its hard to define exactly what should count as spam I’m thinking I’d basically want to filter the same type of stuff that Warpcast hides under “Show more replies.” Like for example this cast has 16 replies but 14 of them are hidden because they're spammy https://warpcast.com/katiewav/0x8466947e
2 replies
0 recast
0 reaction
Victor
@victoreram
Do you have access to user data? The spam filtering might be more effective at a account level than the cast level. If you have both you could assess the spaminess of an account based on some sample of their casts
1 reply
0 recast
1 reaction
Nate Maddrey
@nmadd
yeah I'm using the @airstack.eth Farcaster API makes sense, should be a lot easier to filter out spammy users rather than individual spammy casts. I was thinking of looking at quality of followers too since the spam accounts are probably only followed by other spam accounts https://app.airstack.xyz/api-studio
1 reply
0 recast
1 reaction
Victor
@victoreram
yeah you can probably glean from who spammy users follow. More human-like accounts tend to have some randomness in who they follow. Perhaps also looking into cadence in casting (do they cast on regular intervals). Any behavior that can be programmable could be good indicators of spam
0 reply
0 recast
0 reaction
