Supported Task Types

On this page:

• Supported Task Types
  • Sentiment Analysis
  • Entity Resolution
  • Filtering

The following tasks are available by default with AMPCrowd, though you can always add new task types.

Each task type has a type string to identify the type, a group context to represent shared state across a group of tasks of that type, content to represent the data associated with a single task of that type, and an answer format for returned results.

All task types support an optional instruction field in their group context which may contain HTML-based instructions for the task.

Sentiment Analysis

This task type represents a crowd task to label a text string with its sentiment on a scale from 1 (very negative) to 5 (very positive).

• Type string: 'sa'

• Group Context: none beyond the optional instruction field, e.g.:

  {
      "instruction": "<p>For each tweet, decide whether it is positive,
                          negative, or neutral</p>"
  }
  

• Content: a json dictionary mapping unique ids to text strings, e.g.:

  {
      "aedcf13": "Arsenal won the 4th again!",
      "4abf23d": "Theo Walcott broke the ligament in his knee last season.",
      "cf8f93b": "Lebron James back to Cavs--guess he can't handle the Heat."
  }
  

• Answers: values should be an integer in [1,5] corresponding to:

  • 1: Tweet is very negative.
  • 2: Tweet is somewhat negative.
  • 3: Tweet is neutral.
  • 4: Tweet is somewhat positive.
  • 5: Tweet is very positive.

Entity Resolution

This task type represents a crowd task to determine whether two records refer to the same real-world entity or not.

• Type string: 'er'

• Group Context: In addition to the optional instructions, the group_context consists of the shared schema for each pair of records. For example:

  {
      "instruction": "<p>Decide whether two records in each group refer to
                      the <b>same entity</b>.</p>",
      "fields": ["price", "location"]
  }
  

• Content: Content should consist of pairs of records for entity resolution, specified as a json dictionary with pairs of records mapped to unique ids, e.g:

  {
      "aedcf13": [["5", "LA"], ["6", "Berkeley"]],
      "4abf23d": [["80", "London"], ["80.0", "Londyn"]]
  }
  

• Answers: values should be either 0.0 or 1.0, indicating ‘records do not match’ or ‘records match’, respectively.

Filtering

This task type represents a crowd task to answer simple yes or no questions about individual records.

• Type string: 'ft'

• Group Context: In addition to the optional instructions, the group_context consists of the shared schema of the records in each question. It is similar to the group_context of an entity resolution task. For example:

  {
      "instruction": "<p>Answer the question in each group.</p>",
      "fields": ["Conference", "First Author"]
  }
  

• Content: The content for each point should be a json dictionary consisting of the question title and the values for each data field for the record, e.g:

  {
      "aedcf13": {
          "title": "Is this a paper by Michael Franklin (UC Berkeley)?",
          "record": ["icde", "Michael Franklin"]
      }
  }
  

• Answers: values should be either 0.0 or 1.0, indicating ‘the answer to the question is NO’ or ‘the answer to the question is YES’ respectively.