Last updated

Personal Attack

A personal attack is an insult, an ad hominem attack, an instance of cyberbullying, or any other kind of attack on a participant in a conversation. Instances of detected personal attacks are logged under the abuse section, with the type attribute set to personal_attack.

What Is Not Considered A Personal Attack?

  • An attack on a social group

    • If the target is an ethnic, religious, or racial group, as this is categorized as bigotry (hate speech).
    • Attacks on other social groups may not fall under any of the categories, or they can be marked as contentious (content likely to start disputes).
  • An attack on a fictional character or public figure

    • Criticizing a fictional character or a public figure does not count as a personal attack.
    • An attack on someone currently not part of the conversation. That means anyone who is not a 2nd person ("you"), and is not identified by a mention linked to a user handle (e.g. @user12345).
    • Negative sentiment that does not attack the person directly. For example, "your music is bad" is not a personal attack.

Example

Request:

{
  "language":"en",
  "content":"You are so stupid, nobody likes you here!",
  "settings": 
  {
    "snippets":true, "explain":true
  }
}

Response:

{
	"text": "You are so stupid, nobody likes you here!",
	"abuse": [
		{
			"sentence_index": 0,
			"offset": 0,
			"length": 17,
			"text": "You are so stupid",
			"type": "personal_attack",
			"severity": "medium",
			"explanation": "Calling someone unsatisfactory"
		},
		{
			"sentence_index": 0,
			"offset": 19,
			"length": 16,
			"text": "nobody likes you",
			"type": "personal_attack",
			"severity": "medium",
			"explanation": "Claim of universal dislike"
		}
	]
}

Redefining Default Policy

The default policy of excluding 3rd parties can be modified using the memory module.

Detecting Attacks on Named Parties

In order to detect attacks on people referenced by names, add the following memory setting:

{
  "assign": [
    {
      "if": {
        "features": [
          {
            "index": 1,
            "value": "NOUN"
          },
          {
            "index": 14,
            "value": "NA"
          },
          {
            "index": 22,
            "value": "PERS"
          }
        ]
      },
      "then": {
        "features": [
          {
            "index": 33,
            "value": "DIPA"
          }
        ],
        "hypernym": 123887
      }
    }
  ]
}

For example:

{
  "language": "en",
  "content": "John Smith sucks",
  "settings": 
  {
    "memory":
    {
      "assign": [
      {
        "if": {
          "features": [
            {
              "index": 1,
              "value": "NOUN"
            },
            {
              "index": 14,
              "value": "NA"
            },
            {
              "index": 22,
              "value": "PERS"
            }
          ]
        },
        "then": {
          "features": [
            {
              "index": 33,
              "value": "DIPA"
            }
          ],
          "hypernym": 123887
        }
      } ]
    }
  }
}

Detecting Attacks on all 3rd Parties

In order to detect attacks on all 3rd parties, add the following memory setting:

{
  "assign": [
    {
      "if": {
        "features": [
          {
            "index": 1,
            "value": "NOUN"
          },
          {
            "index": 9,
            "value": "3"
          },
          {
            "index": 22,
            "value": "PERS"
          }
        ]
      },
      "then": {
        "features": [
          {
            "index": 33,
            "value": "DIPA"
          }
        ],
        "hypernym": 123887
      }
    }
  ]
}

For example:

{
  "language": "en",
  "content": "He is an idiot",
  "settings": 
  {
    "memory":
    {
      "assign": [
      {
        "if": {
          "features": [
            {
              "index": 1,
              "value": "NOUN"
            },
            {
              "index": 9,
              "value": "3"
            },
            {
              "index": 22,
              "value": "PERS"
            }
          ]
        },
        "then": {
          "features": [
            {
              "index": 33,
              "value": "DIPA"
            }
          ],
          "hypernym": 123887
        }
      } ]
    }
  }
}