Giters

brightgems / PyABSA

Open & Efficient for Framework for Aspect-based Sentiment Analysis

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

PyABSA - Open & Efficient for Framework for Aspect-based Sentiment Analysis

PyPI - Python Version PyPI Repo Size PyPI_downloads License Gitter

All Contributors

Fast & Low Memory requirement & Enhanced implementation of Local Context Focus.

Build from LC-ABSA / LCF-ABSA / LCF-BERT and LCF-ATEPC.

Provide tutorials of training and usages of ATE and APC models.

PyTorch Implementations (CPU & CUDA supported).

Tips

  • PyABSA use the FindFile to find the target file which means you can specify a dataset/checkpoint by keywords instead of using absolute path. e.g.,
dataset = 'laptop' # instead of './SemEeval/LAPTOP' case doesn't matter
checkpoint = 'lcfs' # any checkpoint whose absoulte path contains lcfs
  • PyABSA use the AutoCuda to support automatic cuda assignment, but you can still set a preferred device.
auto_device=True  # to auto assign a cuda device for training / inference
auto_device='cuda:1'  # to specify a prefered device
auto_device='cpu'  # to specify a prefered device
  • PyABSA support auto label fixing which means you can set the labels to any number (greater than -999), e.g., sentiment labels = {-9. -1, 0, 199}
  • Other features are available to be found

Instruction

If you are willing to support PyABSA project, please star this repository as your contribution.

Installation

Please do not install the version without corresponding release note to avoid installing a test version.

install via pip

To use PyABSA, install the latest version from pip or source code:

pip install -U pyabsa

install via source

git clone https://github.com/yangheng95/PyABSA --depth=1
cd PyABSA 
python setup.py install

Package Overview

pyabsa package root (including all interfaces)
pyabsa.functional recommend interface entry
pyabsa.functional.checkpoint checkpoint manager entry, inference model entry
pyabsa.functional.dataset datasets entry
pyabsa.functional.config predefined config manager
pyabsa.functional.trainer training module, every trainer return a inference model

Quick Start

Aspect Polarity Classification (APC)

1. Import necessary entries

from pyabsa.functional import Trainer
from pyabsa.functional import APCConfigManager
from pyabsa.functional import ABSADatasetList

# Get model list for Bert-based APC models
from pyabsa.functional import APCModelList

# Get model list for Bert-based APC baseline models
# from pyabsa.functional import BERTBaselineAPCModelList 

# Get model list for GloVe-based APC baseline models
# from pyabsa.functional import GloVeAPCModelList

2. Choose a base param config

# Choose a Bert-based APC models param_dict
apc_config_english = APCConfigManager.get_apc_config_english()

# Choose a Bert-based APC baseline models param_dict
# apc_config_english = APCConfigManager.get_apc_config_bert_baseline()

# Choose a GloVe-based APC baseline models param_dict
# apc_config_english = APCConfigManager.get_apc_config_glove()

3. Specify an APC model and alter some hyper-parameters (if necessary)

# Specify a Bert-based APC model
apc_config_english.model = APCModelList.SLIDE_LCFS_BERT

# Specify a Bert-based APC baseline model
# apc_config_english.model = BERTBaselineAPCModelList.ASGCN_BERT

# Specify a GloVe-based APC baseline model
# apc_config_english.model = GloVeAPCModelList.ASGCN

apc_config_english.similarity_threshold = 1
apc_config_english.max_seq_len = 80
apc_config_english.dropout = 0.5
apc_config_english.log_step = 5
apc_config_english.num_epoch = 10
apc_config_english.evaluate_begin = 4
apc_config_english.l2reg = 0.0005
apc_config_english.seed = {1, 2, 3}
apc_config_english.cross_validate_fold = -1

4. Configure runtime setting and running training

dataset_path = ABSADatasetList.SemEval #or set your local dataset
sent_classifier = Trainer(config=apc_config_english,
                          dataset=dataset_path,  # train set and test set will be automatically detected
                          checkpoint_save_mode=1,  # = None to avoid save model
                          auto_device=True  # automatic choose CUDA or CPU
                          )

5. Sentiment inference

# batch inferring_tutorials returns the results, save the result if necessary using save_result=True
inference_dataset = ABSADatasetList.SemEval # or set your local dataset
results = sent_classifier.batch_infer(target_file=inference_dataset,
                                      print_result=True,
                                      save_result=True,
                                      ignore_error=True,
                                      )

6. Sentiment inference output format (情感分类结果示例如下)

Apple is unmatched in  product quality  , aesthetics , craftmanship , and customer service .  
product quality --> Positive  Real: Positive (Correct)
 Apple is unmatched in product quality ,  aesthetics  , craftmanship , and customer service .  
aesthetics --> Positive  Real: Positive (Correct)
 Apple is unmatched in product quality , aesthetics ,  craftmanship  , and customer service .  
craftmanship --> Positive  Real: Positive (Correct)
 Apple is unmatched in product quality , aesthetics , craftmanship , and  customer service  .  
customer service --> Positive  Real: Positive (Correct)
It is a great size and amazing  windows 8  included !  
windows 8 --> Positive  Real: Positive (Correct)
 I do not like too much  Windows 8  .  
Windows 8 --> Negative  Real: Negative (Correct)
Took a long time trying to decide between one with  retina display  and one without .  
retina display --> Neutral  Real: Neutral (Correct)
 It 's so nice that the  battery  last so long and that this machine has the snow lion !  
battery --> Positive  Real: Positive (Correct)
 It 's so nice that the battery last so long and that this machine has the  snow lion  !  
snow lion --> Positive  Real: Positive (Correct)

Check the detailed usages in APC examples directory.

Aspect Term Extraction and Polarity Classification (ATEPC)

1. Import necessary entries

from pyabsa.functional import ATEPCModelList
from pyabsa.functional import Trainer, ATEPCTrainer
from pyabsa.functional import ABSADatasetList
from pyabsa.functional import ATEPCConfigManager

2. Choose a base param config

config = ATEPCConfigManager.get_atepc_config_english()

3. Specify an ATEPC model and alter some hyper-parameters (if necessary)

atepc_config_english = ATEPCConfigManager.get_atepc_config_english()
atepc_config_english.num_epoch = 10
atepc_config_english.evaluate_begin = 4
atepc_config_english.log_step = 100
atepc_config_english.model = ATEPCModelList.LCF_ATEPC

4. Configure runtime setting and running training

laptop14 = ABSADatasetList.Laptop14

aspect_extractor = ATEPCTrainer(config=atepc_config_english, 
                                dataset=laptop14
                                )

5. Aspect term extraction & sentiment inference

from pyabsa import ATEPCCheckpointManager

examples = ['相比较原系列锐度高了不少这一点好与不好大家有争议',
            '这款手机的大小真的很薄，但是颜色不太好看， 总体上我很满意啦。'
            ]
aspect_extractor = ATEPCCheckpointManager.get_aspect_extractor(checkpoint='chinese',
                                                               auto_device=True  # False means load model on CPU
                                                               )

inference_source = pyabsa.ABSADatasetList.SemEval
atepc_result = aspect_extractor.extract_aspect(inference_source=inference_source, 
                                               save_result=True,
                                               print_result=True,  # print the result
                                               pred_sentiment=True,  # Predict the sentiment of extracted aspect terms
                                               )

6. Aspect term extraction & sentiment inference output format (方面抽取及情感分类结果示例如下):

Sentence with predicted labels:
关(O) 键(O) 的(O) 时(O) 候(O) 需(O) 要(O) 表(O) 现(O) 持(O) 续(O) 影(O) 像(O) 的(O) 短(B-ASP) 片(I-ASP) 功(I-ASP) 能(I-ASP) 还(O) 是(O) 很(O) 有(O) 用(O) 的(O)
{'aspect': '短 片 功 能', 'position': '14,15,16,17', 'sentiment': '1'}
Sentence with predicted labels:
相(O) 比(O) 较(O) 原(O) 系(O) 列(O) 锐(B-ASP) 度(I-ASP) 高(O) 了(O) 不(O) 少(O) 这(O) 一(O) 点(O) 好(O) 与(O) 不(O) 好(O) 大(O) 家(O) 有(O) 争(O) 议(O)
{'aspect': '锐 度', 'position': '6,7', 'sentiment': '0'}

Sentence with predicted labels:
It(O) was(O) pleasantly(O) uncrowded(O) ,(O) the(O) service(B-ASP) was(O) delightful(O) ,(O) the(O) garden(B-ASP) adorable(O) ,(O) the(O) food(B-ASP) -LRB-(O) from(O) appetizers(B-ASP) to(O) entrees(B-ASP) -RRB-(O) was(O) delectable(O) .(O)
{'aspect': 'service', 'position': '7', 'sentiment': 'Positive'}
{'aspect': 'garden', 'position': '12', 'sentiment': 'Positive'}
{'aspect': 'food', 'position': '16', 'sentiment': 'Positive'}
{'aspect': 'appetizers', 'position': '19', 'sentiment': 'Positive'}
{'aspect': 'entrees', 'position': '21', 'sentiment': 'Positive'}
Sentence with predicted labels:

Check the detailed usages in ATE examples directory.

Checkpoint

How to get available checkpoints from Google Drive

PyABSA will check the latest available checkpoints before and load the latest checkpoint from Google Drive. To view available checkpoints, you can use the following code and load the checkpoint by name:

from pyabsa import available_checkpoints

checkpoint_map = available_checkpoinbertts()

If you can not access to Google Drive, you can download our checkpoints and load the unzipped checkpoint manually. 如果您无法访问谷歌Drive，您可以下载我们预训练的模型，并手动解压缩并加载模型。 模型下载地址 提取码：ABSA

How to share checkpoints (e.g., checkpoints trained on your custom dataset) with community

For resource limitation, we do not provide diversities of checkpoints, we hope you can share your checkpoints with those who have not enough resource to train their model.

  1. Upload your zipped checkpoint to Google Drive in a shared folder. 123

  2. Get the link of your checkpoint. 123

  3. Register the checkpoint in the checkpoint_map, then make a pull request. We will update the checkpoints index as soon as we can, Thanks for your help!

"checkpoint name": {
        "id": "your checkpoint link",
        "model": "model name",
        "dataset": "trained dataset",
        "description": "trained equipment",
        "version": "used pyabsa version",
        "author": "name (email)"
      }

How to use checkpoints

1. Sentiment inference

1.1 Import necessary entries

import os
from pyabsa import APCCheckpointManager, ABSADatasetList
os.environ['PYTHONIOENCODING'] = 'UTF8'

1.2 Assume the sent_classifier and checkpoint

sentiment_map = {0: 'Negative', 1: 'Neutral', 2: 'Positive', -999: ''}

sent_classifier = APCCheckpointManager.get_sentiment_classifier(checkpoint='dlcf-dca-bert1', #or set your local checkpoint
                                                                auto_device='cuda',  # Use CUDA if available
                                                                sentiment_map=sentiment_map
                                                                )

1.3 Configure inferring setting

# batch inferring_tutorials returns the results, save the result if necessary using save_result=True
inference_datasets = ABSADatasetList.Laptop14 # or set your local dataset
results = sent_classifier.batch_infer(target_file=inference_datasets,
                                      print_result=True,
                                      save_result=True,
                                      ignore_error=True,
                                      )

2. Aspect term extraction & sentiment inference

2.1 Import necessary entries

import os
from pyabsa import ABSADatasetList
from pyabsa import ATEPCCheckpointManager
os.environ['PYTHONIOENCODING'] = 'UTF8'

2.2 Assume the sent_classifier and checkpoint

sentiment_map = {0: 'Negative', 1: "Neutral", 2: 'Positive', -999: ''}

aspect_extractor = ATEPCCheckpointManager.get_aspect_extractor(checkpoint='Laptop14', # or your local checkpoint
                                                               auto_device=True  # False means load model on CPU
                                                               )

2.3 Configure extraction and inferring setting

# inference_dataset = ABSADatasetList.SemEval # or set your local dataset
atepc_result = aspect_extractor.extract_aspect(inference_source=inference_dataset,
                                               save_result=True,
                                               print_result=True,  # print the result
                                               pred_sentiment=True,  # Predict the sentiment of extracted aspect terms
                                               )

3. Train based on checkpoint

3.1 Import necessary entries

from pyabsa.functional import APCCheckpointManager
from pyabsa.functional import Trainer
from pyabsa.functional import APCConfigManager
from pyabsa.functional import ABSADatasetList
from pyabsa.functional import APCModelList

3.2 Choose a base param_dict

apc_config_english = APCConfigManager.get_apc_config_english()

3.3 Specify an APC model and alter some hyper-parameters (if necessary)

apc_config_english.model = APCModelList.SLIDE_LCF_BERT
apc_config_english.evaluate_begin = 2
apc_config_english.similarity_threshold = 1
apc_config_english.max_seq_len = 80
apc_config_english.dropout = 0.5
apc_config_english.log_step = 5
apc_config_english.l2reg = 0.0001
apc_config_english.dynamic_truncate = True
apc_config_english.srd_alignment = True

3.4 Configure checkpoint

# Ensure the corresponding checkpoint of trained model
checkpoint_path = APCCheckpointManager.get_checkpoint('slide-lcf-bert')

3.5 Configure runtime setting and running training

dataset_path = ABSADatasetList.SemEval #or set your local dataset
sent_classifier = Trainer(config=apc_config_english,
                          dataset=dataset_path,
                          from_checkpoint=checkpoint_path,
                          checkpoint_save_mode=1,
                          auto_device=True
                          )

Datasets

More datasets are available at ABSADatasets.

  1. Twitter
  2. Laptop14
  3. Restaurant14
  4. Restaurant15
  5. Restaurant16
  6. Phone
  7. Car
  8. Camera
  9. Notebook
  10. MAMS
  11. Multilingual (The sum of the above datasets.)
  12. TShirt
  13. Television

Basically, you don't have to download the datasets, as the datasets will be downloaded automatically.

Model Support

Except for the following models, we provide a template model involving LCF vec, you can develop your model based on the LCF-APC model template or LCF-ATEPC model template.

ATEPC

  1. LCF-ATEPC
  2. LCF-ATEPC-LARGE (Dual BERT)
  3. FAST-LCF-ATEPC
  4. LCFS-ATEPC
  5. LCFS-ATEPC-LARGE (Dual BERT)
  6. FAST-LCFS-ATEPC
  7. BERT-BASE

APC

Bert-based APC models

  1. SLIDE-LCF-BERT (Faster & Performs Better than LCF/LCFS-BERT)
  2. SLIDE-LCFS-BERT (Faster & Performs Better than LCF/LCFS-BERT)
  3. LCF-BERT (Reimplemented & Enhanced)
  4. LCFS-BERT (Reimplemented & Enhanced)
  5. FAST-LCF-BERT (Faster with slightly performance loss)
  6. FAST_LCFS-BERT (Faster with slightly performance loss)
  7. LCF-DUAL-BERT (Dual BERT)
  8. LCFS-DUAL-BERT (Dual BERT)
  9. BERT-BASE
  10. BERT-SPC
  11. LCA-Net
  12. DLCF-DCA-BERT *

Bert-based APC baseline models

  1. AOA_BERT
  2. ASGCN_BERT
  3. ATAE_LSTM_BERT
  4. Cabasc_BERT
  5. IAN_BERT
  6. LSTM_BERT
  7. MemNet_BERT
  8. MGAN_BERT
  9. RAM_BERT
  10. TD_LSTM_BERT
  11. TC_LSTM_BERT
  12. TNet_LF_BERT

GloVe-based APC baseline models

  1. AOA
  2. ASGCN
  3. ATAE-LSTM
  4. Cabasc
  5. IAN
  6. LSTM
  7. MemNet
  8. MGAN
  9. RAM
  10. TD-LSTM
  11. TD-LSTM
  12. TNet_LF

Contribution

We expect that you can help us improve this project, and your contributions are welcome. You can make a contribution in many ways, including:

  • Share your custom dataset in PyABSA and ABSADatasets
  • Integrates your models in PyABSA. (You can share your models whether it is or not based on PyABSA. if you are interested, we will help you)
  • Raise a bug report while you use PyABSA or review the code (PyABSA is a individual project driven by enthusiasm so your help is needed)
  • Give us some advice about feature design/refactor (You can advise to improve some feature)
  • Correct/Rewrite some error-messages or code comment (The comments are not written by native english speaker, you can help us improve documents)
  • Create an example script in a particular situation (Such as specify a SpaCy model, pretrainedbert type, some hyperparameters)
  • Star this repository to keep it active

Notice

The LCF is a simple and adoptive mechanism proposed for ABSA. Many models based on LCF has been proposed and achieved SOTA performance. Developing your models based on LCF will significantly improve your ABSA models. If you are looking for the original proposal of local context focus, please redirect to the introduction of LCF. If you are looking for the original codes of the LCF-related papers, please redirect to LC-ABSA / LCF-ABSA or LCF-ATEPC.

Acknowledgement

This work build from LC-ABSA/LCF-ABSA and LCF-ATEPC, and other impressive works such as PyTorch-ABSA and LCFS-BERT.

License

MIT

Contributors ✨

Thanks goes to these wonderful people (emoji key):


XuMayi
💻
YangHeng
📆
brtgpy
🔣

This project follows the all-contributors specification. Contributions of any kind welcome!

About

Open & Efficient for Framework for Aspect-based Sentiment Analysis

License:MIT License

Languages

Language:Python 100.0%