Spaces:
Running
Running
File size: 7,510 Bytes
38fb4a1 5ae1418 bf1042d 5ae1418 b42b884 5ae1418 e50b20c b42b884 5ae1418 433d10a 71069d7 667fe9d 85ac990 71069d7 e1645d7 71069d7 d29d6fe 71069d7 e3095cd 419453c 71069d7 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 |
---
title: Sentiment Analysis
emoji: 🤗
colorFrom: blue
colorTo: green
pinned: false
sdk: gradio
python_version: 3.11
app_file: app/gui.py
datasets:
- mrshu/amazonreviews
- stanfordnlp/sentiment140
- stanfordnlp/imdb
- Sp1786/multiclass-sentiment-analysis-dataset
models:
- spacy/en_core_web_sm
---
# Sentiment Analysis [![Hugging Face Spaces](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue)](https://huggingface.co/spaces/Tymec/sentiment-analysis)
### Table of Contents
- [Description](#description)
- [Installation](#installation)
- [Prerequisites](#prerequisites)
- [Usage](#usage)
- [Predict](#predict)
- [GUI](#gui)
- [Training](#training)
- [Evaluation](#evaluation)
- [Options](#options)
- [Datasets](#datasets)
- [Vectorizers](#vectorizers)
- [Environment Variables](#environment-variables)
- [Implementation](#implementation)
- [Architecture](#architecture)
- [Pre-trained Models](#pre-trained-models)
- [License](#license)
## Description
This is a simple sentiment analysis model written in Python, designed to predict whether the provided text has a positive or negative sentiment. The project comes with both a graphical user interface and a command-line interface. While training the model, the user can choose from a couple of datasets to train the model on and then evaluate the trained model on another dataset. Once the model is trained, it can be used to predict the sentiment of any text with the help of the GUI or CLI.
## Installation
Clone the repository and once inside the directory, run the following command to install the dependencies:
```bash
python -m pip install -r requirements.txt
```
Ensure that you have **at least** one dataset downloaded and placed in the data directory before running `train`.
For `evaluate`, you will need the `test` dataset. See [Datasets](#datasets) for more information.
The project comes with pre-trained models that can be used for prediction. See [Pre-trained Models](#pre-trained-models) for more information.
### Prerequisites
- Python 3.11+
## Usage
To see the available commands and options, run:
```bash
python -m app --help
```
<!-- Image of the output -->
### Predict
To perform sentiment analysis on a given text, run the following command:
```bash
python -m app predict --model <model> I love this movie
```
where `<model>` is the path to the trained model.
Alternatively, you can pipe the text into the command:
```bash
echo "I love this movie" | python -m app predict --model <model>
```
<!-- Image of the output -->
### GUI
To launch the GUI, run the following command:
```bash
python -m app gui --model <model>
```
where `<model>` is the path to the trained model. Add the `--share` flag to create a publicly accessible link.
After running the command, open the link from the terminal in your browser to access the GUI.
<!-- Image of the output -->
<!-- Image of the GUI -->
### Training
Before training the model, ensure that the specified dataset is downloaded and can be accessed at its respective path. To train the model, run the following command:
```bash
python -m app train --dataset <dataset> {options}
```
where `<dataset>` is the name of the dataset to train the model on. For available datasets, see [Datasets](#datasets).
The trained model will be exported to the models directory.
To see all available options, run:
```bash
python -m app train --help
```
<!-- Image of the output -->
### Evaluation
Once the model is trained, you can evaluate it on a different dataset by running the following command:
```bash
python -m app evaluate --model <model>
```
where `<model>` is the path to the trained model. For available datasets, see [Datasets](#datasets).
To see all available options, run:
```bash
python -m app evaluate --help
```
<!-- Image of the output -->
## Options
### Datasets
| Option | Path | Notes | Dataset |
| --- | --- | --- | --- |
| sentiment140 | `data/sentiment140.csv` | | [Twitter Sentiment Analysis](https://www.kaggle.com/kazanova/sentiment140) |
| amazonreviews | `data/amazonreviews.bz2` | only train is used | [Amazon Product Reviews](https://www.kaggle.com/bittlingmayer/amazonreviews) |
| imdb50k | `data/imdb50k.csv` | | [IMDB Movie Reviews](https://www.kaggle.com/lakshmi25npathi/imdb-dataset-of-50k-movie-reviews) |
| test | `data/test.csv` | required for `evaluate` | [Multiclass Sentiment Analysis](https://huggingface.co/datasets/Sp1786/multiclass-sentiment-analysis-dataset) |
#### Used for text preprocessing
- [Slang Map](Https://www.kaggle.com/code/nmaguette/up-to-date-list-of-slangs-for-text-preprocessing)
### Vectorizers
| Option | Description | When to Use |
| --- | --- | --- |
| `count` | Count Vectorizer | When the frequency of words is important |
| `tfidf` | TF-IDF Vectorizer | When the importance of words is important |
| `hashing` | Hashing Vectorizer | When memory is a concern |
### Environment Variables
The following environment variables can be set to customize the behavior of the application:
| Name | Description | Default |
| --- | --- | --- |
| `MODEL_DIR` | the directory where the trained models are stored | `models` |
| `DATA_DIR` | the directory where the datasets are stored | `data` |
| `CACHE_DIR` | the directory where cached files are stored | `.cache` |
## Implementation
### Architecture
The input text is first preprocessed and tokenized using `re` and `spaCy` where:
- The text is cleaned up by removing any HTML tags and converting emojis to text
- Stop words and punctuation are removed
- URLs, email addresses and numbers are removed
- Words are converted to lowercase
- Lemmatization is performed (words are converted to their base form based on the surrounding context)
After tokenization, feature extraction is performed on the tokens using the chosen vectorizer. Each vectorizer has its own advantages and disadvantages, and the choice of vectorizer can affect the speed and accuracy of the model (see [Vectorizers](#vectorizers)). The extracted features are then passed to the classifier which predicts the class which in this case is the sentiment of the text. Both the vectorizer and classifier are trained on the specified dataset.
```mermaid
%%{ init : { "flowchart" : { "curve" : "monotoneX" }}}%%
graph LR
START:::hidden --> |text|Preprocessing
subgraph Preprocessing
direction TB
A[Tokenizer]
B1[HashingVectorizer]
B2[CountVectorizer]
B3[TfidfVectorizer]
A --> B1
A --> |tokens|B2
A --> B3
B1 --> C1:::hidden
B2 --> C2:::hidden
B3 --> C3:::hidden
end
Preprocessing --> |features|Classification
subgraph Classification
direction LR
D1[LogisticRegression]
D2[LinearSVC]
end
Classification --> |sentiment|END:::hidden
classDef hidden display: none;
```
### Pre-trained Models
The following pre-trained models are available for use:
| Dataset | Vectorizer | Classifier | Features | Accuracy on test | Accuracy on self | Model |
| --- | --- | --- | --- | --- | --- | --- |
| `imdb50k` | `tfidf` | `LinearRegression` | 20 000 | 59.59% ± 0.73% | 74.08% ± 0.14% | [Here](models/imdb50k_tfidf_ft20000.pkl) |
| `sentiment140` | `tfidf` | `LinearRegression` | 20 000 | 52.81% ± 0.03% | ❌ | [Here](models/sentiment140_tfidf_ft20000.pkl) |
| `amazonreviews` | `tfidf` | `LinearRegression` | 20 000 | ❌ | ❌ | [Here](models/amazonreviews_tfidf_ft1048576.pkl) |
## License
Distributed under the MIT License. See [LICENSE](LICENSE) for more information.
|