Need better way to distinguish offensive content
Consider the following two tweets. The first is benign. The second is offensive. How can I know that progamatically? They have the same ReviewRecommended, the same Category3 score, low Cat1 and Cat2 scores, and neither have any Terms flagged. This is useless.
{
"OriginalText": "Wanna win tix to our @RoughTradeNYC show NEXT MON 5/14? @Thrillcall's got you covered. Enter to win here >… https://t.co/s0KT4JDCB1",
"NormalizedText": "Wanna win tix to our @ RoughTradeNYC show NEXT MON 5/ 14? @ Thrillcall' s got you covered. Enter to win here & gt; … https://t.co/s0KT4JDCB1",
"Misrepresentation": null,
"Classification":
{
"ReviewRecommended": true,
"Category1":
{
"Score": 0.001084162387996912
},
"Category2":
{
"Score": 0.13056860864162445
},
"Category3":
{
"Score": 0.98799997568130493
}
},
"Language": "eng",
"Terms": null,
"Status":
{
"Code": 3000,
"Description": "OK",
"Exception": null
},
"TrackingId": "b1f79713-f84f-47b5-b4d8-0a0d0aec3d73"
}
compare to:
{
"OriginalText": "@omgimwigs @ me bro fckin @ me 😤🔫🙅🏻",
"NormalizedText": "@ omgimwigs @ me bro fckin @ me 😤🔫🙅🏻",
"Misrepresentation": null,
"Classification":
{
"ReviewRecommended": true,
"Category1":
{
"Score": 0.28458982706069946
},
"Category2":
{
"Score": 0.19746741652488708
},
"Category3":
{
"Score": 0.98799997568130493
}
},
"Language": "eng",
"Terms": null,
"Status":
{
"Code": 3000,
"Description": "OK",
"Exception": null
},
"TrackingId": "30bf4c39-47ae-4047-8ebb-863486bbabab"
}
Jeff Lit
shared this idea
2 comments
-
Jeff, We are connected via email so I am resolving this thread. As soon as our new update is in production, I will let you know via email.
-
Jeff, We are looking into this in terms of improving the model. An interim workaround would be to use the List Management API to create your custom "allow" or "Block" term lists beyond the default/built-in lists. Will send you the links via email. - Sanjeev
