README.md

2023 Google Pixel User Sentiment Monitoring

Links

• 2023 Progress Reports @ Pixel User Sentiment Monitoring Sync

1. Crawler

• 🗓 reported date:
  • 2022/09/26
  • 2022/10/12
• 🌲 branch: crawler
• 🗂️ The crawler to crawl the data from the Google Pixel forum.
• 🔩 toolkits:
  • framework: scrapy.

2. Pixel ABSA TestSet (PATS)

• 🗓 reported date:
  • 2023/08/04
  • 2023/08/17
• 🌲 branch: pats
• 🗂️ Labeled data of the crawled data.
• 🔩 toolkits:
  • labeling platform: Label Studio
• The scripts inside convert the labeled data from label-studio format to the format that can be used by (1) bert-based models: RobertaABSA model and (2) text-to-text models: LLaMA.

3. BERT-Series: A Simple ABSA Baseline

• 🗓 reported date:
  • 2023/02/17
• 🌲 branch: bert_baseline
• 🗂️ A training and testing script for running BERT-based ATE+ALSC pipeline for simple aspect-based sentiment analysis.
• 🔩 toolkits:
  • framework: PyTorch Lightning + miracleyoo/pytorch-lightning-template

4. LLM-Series: ABSA with ChatGPT

• 🗓 reported date:
  • 2023/04/14
• 🌲 branch: N/A
• 🔩 toolkits:
  • data analysis tool: tableau
• 🗂️ content:
  • visualization: Pixel Sentiment Monitoring gpt baseline @ Tableau Public
  • report: 2023/04/14 Progress Report p.6-11

5. LLM-Series: Senti-LLaMA Experiments [to be updated]

• 🗓 reported date:
  • 2023/07/07 ~ 2023/09/15
• 🌲 branch: senti_llama
• 🗂️ content:
  • training script and hyperparam configs for finetuning LLaMA 2 (-chat) on sentiment datasets
  • sentiment-LLaMA weights
  • sentiment-LLaMA input-format converted datasets
    • yelp, imdb, ... [to be updated]
• 🔩 toolkits:
  • training streamline: OpenAccess-AI-Collective/axolotl
  • attention mechanism: Dao-AILab/flash-attention
  • efficiency tool: microsoft/DeepSpeed

6. Dialogue ABSA

• 🗓 reported date:
  • 2023/09/15
  • 2023/06/09
• 🌲 repo: Nana2929/dialogue-absa
• 🗂️ content:
  • We test the dialogue absa dataset DiaASQ with 3 models. See the repo for more details.
    • analysis: 0609report, 0915 report
    • statistics
  • 🌲 branch: data: conversion scripts for generative model (in our case, LLaMA-2) training.
  • 🌲 branch: gpt: api calling scripts for ChatGPT.
  • 🌲 branch: llama: metrics scripts for generative model (in our case, all) inference.
  • 🌲 branch: train: training scripts for T5.
• 🔩 toolkits:
  • framework: PyTorch Lightning + miracleyoo/pytorch-lightning-template