mBart / README.md
merqsous's picture
Update README.md
e5dbc95 verified
metadata
license: apache-2.0
language:
  - en
baseline: facebook_BART
pipeline_tag: text-classification
tags:
  - mobile_application
  - user_reviews
widget:
  - text: >-
      I can see the intentions behind this app but the execution feels very much
      lackluster. The quizzes are full of grammar errors and the questions feel
      very surface level. The quizzes also seem to be aimed at current short
      term issues versus establishing a baseline and working on keeping track of
      daily changes. I just see how helpful the app would be to someone to use
      everyday if there is no baseline established.
    example_title: Example 1
  - text: >-
      The 'affirmations' are not like the ones advertised. Possibly just AI
      made... But I don't know. Plus it says 'in app purchases' but it isn't
      stated that you have to pay for the app. Not worth it. Not even worth
      keeping for the entire 3 day trial. Sorry. I gave it an extra star because
      at least it looks pretty and I liked that you could (sort of) schedule
      when the affirmations arrived. It is a lovely idea but I don't feel it
      stuck the landing and I definitely don't want to pay for it.
    example_title: Example 2
  - text: >-
      I want to like and use this app. However, it's extremely buggy. It just
      displayed a loading spinner for 3 minutes before finally opening. The
      app's busy spinner is the same as their logo, which is appropriate -
      you'll see that spinner a lot. Also, the UI is a strange mixture of
      polished yet confusing. Everything looks nice, but I still have a hard
      time navigating around. All that said, the mindfulness meditation
      exercises leave me feeling fantastic. They make the app worthwhile to me.
    example_title: Example 3
  - text: >-
      I just experienced biofeedback therapy and learned that I don't breathe
      from my diaphragm most of the time. In looking for a guided breathing app,
      I finally discovered this simple gem. You just breathe with the ball as it
      expands and releases at a gentle pace. The pace is adjustable. I'm not
      there yet but I can feel and breathe the difference in a few sessions. The
      developer's response to a minor question was answered literally overnight:
      I wrote at 1 a.m. and had an answer by 8 a.m.. This guided breathing app
      would be great for people who, like me, don't like meditation guides with
      somber voices, or struggle with simple belly breathing.
    example_title: Example 4

Model Name

Model Description

This multi-label classification model is a fine-tuned version of Facebook's BART model specifically tailored for analyzing mobile application user reviews. It can classify reviews into 38 distinct classes, covering various aspects like usability, functionality, pricing, and more. This model aims to assist developers and marketers in gaining insights from user feedback efficiently.

Model Details

  • Architecture: The model is based on the BART (Bidirectional and Auto-Regressive Transformers) architecture, renowned for its effectiveness in natural language understanding tasks.
  • Training Data: Fine-tuned on a dataset comprising user reviews from multiple mobile applications, collected from various app stores. The dataset includes over 4k user reviews categorized into 38 classes.
  • Training Procedure: The model was trained using a multi-label classification approach, with a focus on maximizing the accuracy across all categories. Training involved adjusting the learning rate and using a batch size optimized for GPU capabilities.

Example Usage

from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained("merqsous/mBart")
tokenizer = AutoTokenizer.from_pretrained("merqsous/mBart")

inputs = tokenizer("Example input text here", return_tensors="pt")
outputs = model(**inputs)