Skip to content

Release v0.1.1

Latest
Compare
Choose a tag to compare
@wshi991201 wshi991201 released this 04 Apr 03:43
· 4 commits to master since this release

Release Note: New Supported Query Types

We're excited to introduce new query types that enhance interactive ranking and sentiment classification capabilities based on the platform NEXT created by Neuromatch. Below is a summary of the newly supported query types and their features.


A Rank B

Overview:
Participants are presented with a pool of A items and must select and rank B items (with B < A) based on a designated anchor item.

User Interface:

  • Red Box: Displays the full pool of items.
  • Green Box: Hosts the B items selected for ranking.
  • Interactions:
    • Clicking an item in the red box moves it to the green box.
    • Clicking an item in the green box returns it to the red box.
    • Drag and drop within the green box allows for rearrangement of the selected items.
  • Submission Rule:
    • Responses can only be submitted when exactly B items are present in the green box.

Dynamic Sampling Algorithms:

  • Random Sampling:
    • Randomly selects A+1 items from the user-provided targets.
    • Configurable parameters include A, B, and the target set.
  • Info-Tuple Sampling:
    • Dynamically updates an embedding based on real-time responses.
    • Selects the tuple of items expected to yield the highest information gain.
    • Offers additional parameters such as burn-in iterations, total iterations, and down-sampling rate, adjustable via the YAML template.

Binary Sentiment Word Classification (BSWC)

Overview:
This query type focuses on sentiment classification of words. Participants are tasked with either:

  • Rank One: Selecting the most positive word from a pool of n words.
  • Rank N: Ranking words from the most positive to the most negative.

Query Interfaces:

  • Rank One Interface:
    • A central box displays the full pool of words.
    • Clicking a word highlights it to indicate the participant's choice.
  • Rank N Interface:
    • Similar to Rank One with a central display of words.
    • Clicking highlights the selected word.
    • Submission is enabled only when exactly one word is highlighted.

Static Sampling Algorithm:

  • Read CSV:
    • Queries are loaded from a CSV file, with each row representing one query.
    • The first number_of_queries rows are truncated, and the remaining rows are presented sequentially as queries.

These enhancements offer robust options for both dynamic and static query sampling, supporting a variety of research and data collection needs. For further configuration details, please consult the respective YAML template documentation.

Happy Ranking!