Hugging Face's logo

Datasets:
stanfordnlp
/
craigslist_bargains

Tasks:
Text Generation
Fill-Mask
Sub-tasks:
dialogue-modeling
Languages:
English
Size:
1K<n<10K
ArXiv:
License:
Dataset card Files Community
4
craigslist_bargains / README.md
albertvillanova's picture
albertvillanova HF staff
Reorder split names (#1)
a3d49d1
preview code
|
raw
history blame
9.53 kB
metadata
annotations_creators:
  - machine-generated
language_creators:
  - crowdsourced
language:
  - en
license:
  - unknown
multilinguality:
  - monolingual
size_categories:
  - 1K<n<10K
source_datasets:
  - original
task_categories:
  - text-generation
  - fill-mask
task_ids:
  - dialogue-modeling
paperswithcode_id: craigslistbargains
pretty_name: CraigslistBargains
dataset_info:
  features:
    - name: agent_info
      sequence:
        - name: Bottomline
          dtype: string
        - name: Role
          dtype: string
        - name: Target
          dtype: float32
    - name: agent_turn
      sequence: int32
    - name: dialogue_acts
      sequence:
        - name: intent
          dtype: string
        - name: price
          dtype: float32
    - name: utterance
      sequence: string
    - name: items
      sequence:
        - name: Category
          dtype: string
        - name: Images
          dtype: string
        - name: Price
          dtype: float32
        - name: Description
          dtype: string
        - name: Title
          dtype: string
  splits:
    - name: train
      num_bytes: 8538836
      num_examples: 5247
    - name: test
      num_bytes: 1353933
      num_examples: 838
    - name: validation
      num_bytes: 966032
      num_examples: 597
  download_size: 25373618
  dataset_size: 10858801

Dataset Card for CraigslistBargains

Table of Contents

Dataset Description

Dataset Summary

We study negotiation dialogues where two agents, a buyer and a seller, negotiate over the price of an time for sale. We collected a dataset of more than 6K negotiation dialogues over multiple categories of products scraped from Craigslist. Our goal is to develop an agent that negotiates with humans through such conversations. The challenge is to handle both the negotiation strategy and the rich language for bargaining. To this end, we develop a modular framework which separates strategy learning from language generation. Specifically, we learn strategies in a coarse dialogue act space and instantiate that into utterances conditioned on dialogue history.

Supported Tasks and Leaderboards

Languages

This dataset is English

Dataset Structure

Data Instances 

{
  'agent_info': {
    'Bottomline': 
      [
        'None', 
        'None'
      ],
    'Role': 
      [
        'buyer', 
        'seller'
      ],
    'Target': 
      [
        7.0, 
        10.0
      ]
  },
  'agent_turn': 
    [
      0, 
      1, 
      ...
    ],
  'dialogue_acts': {
    'intent': 
      [
        'init-price',
        'unknown',
        ...
      ],
    'price': 
      [
        5.0, 
        -1.0, 
        ...
        ]
    },
  'items': {
    'Category': 
      [
        'phone', 
        'phone'
      ],
    'Description': 
      [
        'Charge two devices simultaneously on the go..., 
        ...
      ],
    'Images': 
      [
        'phone/6149527852_0.jpg', 
        'phone/6149527852_0.jpg'
      ],
    'Price': 
      [
        10.0, 
        10.0
      ],
    'Title': 
      [
        'Verizon Car Charger with Dual Output Micro USB and ...', 
        ...
      ]
    },
  'utterance': 
    [
      'Hi, not sure if the charger would work for my car...'
      'It will work...',
      ...
    ]
}

Data Fields

  • agent_info: Information about each of the agents taking part in the dialogue
    • Bottomline: TBD
    • Role: Whether the agent is buyer or seller
    • Target: Target price that the buyer/seller wants to hit in the negotiation
  • agent_turn: Agent taking the current turn in the dialogue (int index corresponding to Role above)
  • dialogue_acts: Rules-based information about the strategy of each agent for each turn
    • intent: The intent of the agent at the particular turn (offer, accept, etc.)
    • price: The current item price associated with the intent and turn in the bargaining process. Default value for missing: (-1)
  • items: Information about the item the agents are bargaining for. Note that there is an elembet for each of the fields below for each agent
    • Category: Category of the item
    • Description: Description(s) of the item
    • Images: (comma delimited) strings of image names of the item
    • Price: Price(s) of the item. Default value for missing: (-1)
    • Title: Title(s) of the item
  • utterance: Utterance for each turn in the dialogue, corresponding to the agent in agent_turns. The utterance may be an empty string ('') for some turns if multiple dialogue acts take place after an utterance (e.g. there are often multiple dialogue acts associated with the closing of the bargaining process after all utterances have completed to describe the conclusion of the bargaining).

Data Splits

This dataset contains three splits, train, validation and test. Note that test is not provided with dialogue_acts information as described above. To ensure schema consistency across dataset splits, the dialogue_acts field in the test split is populated with the default values: {"price": -1.0, "intent": ""}

The counts of examples in each split are as follows:

| | Train | Valid | Test | | Input Examples | 5247 | 597 | 838 | | Average Dialogue Length | 9.14 | 9.17 | 9.24 |

Note that

Dataset Creation

From the source paper for this dataset:

To generate the negotiation scenarios, we scraped postings on sfbay.craigslist.org from the 6 most popular categories (housing, furniture, cars, bikes, phones, and electronics). Each posting produces three scenarios with the buyer’s target prices at 0.5x, 0.7x and 0.9x of the listing price. Statistics of the scenarios are shown in Table 2. We collected 6682 human-human dialogues on AMT using the interface shown in Appendix A Figure 2. The dataset statistics in Table 3 show that CRAIGSLISTBARGAIN has longer dialogues and more diverse utterances compared to prior datasets. Furthermore, workers were encouraged to embellish the item and negotiate side offers such as free delivery or pick-up. This highly relatable scenario leads to richer dialogues such as the one shown in Table 1. We also observed various persuasion techniques listed in Table 4 such as embellishment,

Curation Rationale

See Dataset Creation

Source Data

See Dataset Creation

Initial Data Collection and Normalization

See Dataset Creation

Who are the source language producers?

See Dataset Creation

Annotations

If the dataset contains annotations which are not part of the initial data collection, describe them in the following paragraphs.

Annotation process

Annotations for the dialogue_acts in train and test were generated via a rules-based system which can be found in this script

Who are the annotators?

[More Information Needed]

Personal and Sensitive Information

[More Information Needed]

Considerations for Using the Data

[More Information Needed]

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

[More Information Needed]

Additional Information

[More Information Needed]

Dataset Curators

He He and Derek Chen and Anusha Balakrishnan and Percy Liang Computer Science Department, Stanford University {hehe,derekchen14,anusha,pliang}@cs.stanford.edu

The work through which this data was produced was supported by DARPA Communicating with Computers (CwC) program under ARO prime contract no. W911NF15-1-0462

Licensing Information

[More Information Needed]

Citation Information 

@misc{he2018decoupling,
    title={Decoupling Strategy and Generation in Negotiation Dialogues},
    author={He He and Derek Chen and Anusha Balakrishnan and Percy Liang},
    year={2018},
    eprint={1808.09637},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

Contributions

Thanks to @ZacharySBrown for adding this dataset.