Home PageStudio

Articles vs. Conversations

Suggest Edits

Articles and Conversations

The API handles articles and conversations differently. In fact, from the AI perspective articles and conversations are so distinguished, that it employs completely different models to handle each of them. For this reason, the Pipeline API needs to identify the input type.

Passing input_type

In any case when you know the type of input you are processing, please make sure to pass it in the payload for the pipeline request.

Native

If you work with a native library, set the input_type field of the payload to either article or conversation.

curl -X POST \
'https://api.oneai.com/api/v0/pipeline' \
-H 'accept: application/json' \
-H 'Content-Type: application/json' \
-H 'api-key: <YOUR-API-KEY-HERE>' \
-d '{
    "input": [{ "speaker": "Josh", "utterance": "Do you like math?", "timestamp": "0:00" }, { "speaker": "Mellisa", "utterance": "I like it but I'\''m not so good at it", "timestamp": "0:00" }, { "speaker": "Josh", "utterance": "...", "timestamp": "..." }],
    "input_type": "conversation",
    "steps": [
      {
        "skill": "summarize"
      }   
    ]
}'

SDK

When working with the SDK, the default input_type is article, unless you pass a Conversation object to the pipeline, which then sets the input_type to conversation behind the scenes.

const { OneAI } = require("oneai");

const oneai = new OneAI("<YOUR-API-KEY-HERE>");
const conversation = new oneai.Conversation(
  [
    { "speaker": "Josh", "utterance": "Do you like math?", "timestamp": "0:00" },
    { "speaker": "Mellisa", "utterance": "I like it but I'm not so good at it", "timestamp": "0:02" },
    ]
);

const pipeline = new oneai.Pipeline(
    oneai.skills.splitByTopic(),
);

pipeline.run(conversation).then(console.log);

import oneai

oneai.api_key = "<YOUR-API-KEY-HERE>"
pipeline = oneai.Pipeline(
  steps = [
        oneai.skills.SplitByTopic(),
  ]
)

conversation = oneai.Conversation(
  [
    oneai.Utterance(speaker="Josh", utterance="Do you like math?"), 
    oneai.Utterance(speaker="Mellisa", utterance="I like it but I'm not so good at it"),  
  ]
)

output = pipeline.run(conversation)

Conversation Format

The API accepts conversations in the following format:

[
    { 
    "speaker": "Josh", 
    "utterance": "Do you like math?", 
    "timestamp": "0:00" 
  },
  { 
    "speaker": "Mellisa", 
    "utterance": "I like it but I'm not so good at it", 
    "timestamp": "0:02" 
  },
]

As you can see, a conversation is an array of utterances, each containingspeaker, timestamp and the utterance itself (i.e. what was actually said by the speaker at the timestamp.

Conversation Parsing

To help you format the conversation correctly, our SDKs contain a utility to help just with that

const { OneAI } = require("oneai");

const oneai = new OneAI("<YOUR-API-KEY-HERE>");

const text = `
Josh 00:00
Do you like math?

Mellisa 00:02
I like it but I'm not so good at it.
`;

const conversation = oneai.Conversation.parse(text)

import oneai

text = """
Josh 00:00
Do you like math?

Mellisa 00:02
I like it but I'm not so good at it.
"""

oneai.api_key = "<YOUR-API-KEY-HERE>"

conversation = oneai.Conversation.parse(text)

Updated 5 months ago