Senta is an open-source sentiment analysis system by Baidu’s technical team. Sentiment analysis aims to automatically identify and extract subjective information such as tendencies, positions, evaluations and opinions from texts. It includes a variety of tasks, such as sentence level emotion classification, evaluation object level emotion classification, opinion extraction, emotion classification and so on. Sentiment analysis is an important research direction of artificial intelligence and has high academic value. At the same time, sentiment analysis has important applications in consumer decision-making, public opinion analysis, personalized recommendation and other fields.

In order to facilitate R&D personnel and business partners to share the leading emotion analysis technology, Baidu has open source the SKEP-based emotion pre-training code and the Chinese-English emotion pre-training model in Senta. Users only need a few lines of code to realize the SkeP-based emotion pre-training and model prediction function.

Usage

1 Environmental preparation

PaddlePaddle installation

This project depends on PaddlePaddle 1.6.3. After PaddlePaddle is installed, dynamic library paths such as CUDA, cuDNN, NCCL2 need to be added to environment variable LD_LIBRARY_PATH in time. Otherwise, related library errors will be reported during training.

Detailed installation documentation:
https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/pip/linux-pip.html

pip installation is recommended:

python -m pip install paddlepaddle-gpu==1.6.3.post107-i https://mirror.baidu.com/pypi/simple

senta project python package dependencies

Support for Python 3 version 3.7; The other python package dependencies in the project are listed in the requirements.txt file in the root directory and are installed with the following command:

python -m pip install -r requirements.txt

Environment variable added

Modify the environment variables in./env.sh, including the python, CUDA, cuDNN, NCCL2, PaddlePaddle related environment variables.

PaddlePaddle Environment variable description:
https://www.paddlepaddle.org.cn/documentation/docs/zh/1.6/flags_cn.html < / p >

2 Installation project

Senta warehouse supports both pip installation and source installation. PaddlePaddle needs to be installed before using pip or source installation.

pip installation

python -m pip install Senta

Source code installation

git clone  https://github.com/baidu/Senta.git
cd Senta
python -m pip install .

3 Usage Method

from senta import Senta

my_senta = Senta()

# Get the currently supported emotion pre-training model, We open SKEP models initialized with ERNIE 1.0 large(Chinese), ERNIE 2.0 large(English), and RoBERTa large(English) 
print(my_senta.get_support_model()) # ["ernie_1.0_skep_large_ch", "ernie_2.0_skep_large_en", "roberta_skep_large_en"]

# Get currently supported prediction tasks 
print(my_senta.get_support_task()) # ["sentiment_classify", "aspect_sentiment_classify",  "extraction"]

# Select whether to use gpu
use_cuda = True # Set True or False

# Predicting Chinese sentence-level emotion classification task 
my_senta.init_model(model_class="ernie_1.0_skep_large_ch", task="sentiment_classify", use_cuda=use_cuda)
texts = [" Sun Yat-sen University is the first institution of Learning in Lingnan "]
result = my_senta.predict(texts)
print(result)
# Predicting emotion classification tasks at the Chinese evaluation object level 
my_senta.init_model(model_class="ernie_1.0_skep_large_ch", task="aspect_sentiment_classify", use_cuda=use_cuda)
texts = [" Baidu is a high-tech company "]
aspects = [" Baidu "]
result = my_senta.predict(texts, aspects)
print(result)

# Predict Chinese opinion extraction task 
my_senta.init_model(model_class="ernie_1.0_skep_large_ch", task="extraction", use_cuda=use_cuda)
texts = [" Tang Jia Sanshao, real name Zhang Wei. "]
result = my_senta.predict(texts, aspects)
print(result)

# Predicting English sentence-level sentiment classification Tasks (based on SKEP-ERNIE2.0 model) 
my_senta.init_model(model_class="ernie_2.0_skep_large_en", task="sentiment_classify", use_cuda=use_cuda)
texts = ["a sometimes tedious film ."]
result = my_senta.predict(texts)
print(result)

# Predicting emotion classification tasks at the English Evaluation object level (based on SKEP-ERNIE2.0 model) 
my_senta.init_model(model_class="ernie_2.0_skep_large_en", task="aspect_sentiment_classify", use_cuda=use_cuda)
texts = ["I love the operating system and the preloaded software."]
aspects = ["operating system"]
result = my_senta.predict(texts, aspects)
print(result)

# Predicting English Opinion Extraction Tasks (based on SKEP-ERNIE2.0 model) 
my_senta.init_model(model_class="ernie_2.0_skep_large_en", task="extraction", use_cuda=use_cuda)
texts = ["The JCC would be very pleased to welcome your organization as a corporate sponsor  ."]
result = my_senta.predict(texts)
print(result)

# Predicting English sentence-level sentiment classification tasks (based on SKEP-RoBERTa model) 
my_senta.init_model(model_class="roberta_skep_large_en",  task="sentiment_classify", use_cuda=use_cuda)
texts = ["a sometimes tedious film ."]
result = my_senta.predict(texts)
print(result)

# Predicting emotion classification tasks at the English evaluation object level (based on SKEP-RoBERTa model) 
my_senta.init_model(model_class="roberta_skep_large_en",  task="aspect_sentiment_classify", use_cuda=use_cuda)
texts = ["I love the operating system and the preloaded software."]
aspects = ["operating system"]
result = my_senta.predict(texts, aspects)
print(result)

# Predicting English opinion Extraction Tasks (based on SKEP-RoBERTa model) 
my_senta.init_model(model_class="roberta_skep_large_en",  task="extraction", use_cuda=use_cuda)
texts = ["The JCC would be very pleased to welcome your organization as a corporate sponsor  ."]
result = my_senta.predict(texts)
print(result)

Specific effect

SKEP uses affective knowledge to enhance the pre-training model, and comprehensively outperforms SOTA in 14 typical tasks of Chinese-English affective analysis, with an average improvement of about 2% compared with the original SOTA. The specific results are shown in the table below:

Sentiment analysis system based on SKEP algorithm插图1

You can read more on your own.

资源下载此资源为免费资源立即下载

Telegram:@John_Software

collect(0) Like (0)

Disclaimer: This article is published by a third party and represents the views of the author only and has nothing to do with this website. This site does not make any guarantee or commitment to the authenticity, completeness and timeliness of this article and all or part of its content, please readers for reference only, and please verify the relevant content. The publication or republication of articles by this website for the purpose of conveying more information does not mean that it endorses its views or confirms its description, nor does it mean that this website is responsible for its authenticity.

Ictcoder Free source code Sentiment analysis system based on SKEP algorithm https://ictcoder.com/kyym/sentiment-analysis-system-based-on-skep-algorithm.html

lllll

Share free open-source source code

Previous article： Mockcat, a front - and back-end separate development tool set

Next article： Linux Foundation Open Source: real-time 3D engine for multiple platforms

Q&A

What is the delivery method?

1, automatic: after taking the photo, click the (download) link to download; 2. Manual: After taking the photo, contact the seller to issue it or contact the official to find the developer to ship.

View details

How long is the trading cycle?

1, the default transaction cycle of the source code: manual delivery of goods for 1-3 days, and the user payment amount will enter the platform guarantee until the completion of the transaction or 3-7 days can be issued, in case of disputes indefinitely extend the collection amount until the dispute is resolved or refunded!

View details

Matters needing attention

1. Heptalon will permanently archive the process of trading between the two parties and the snapshots of the traded goods to ensure that the transaction is true, effective and safe! 2, Seven PAWS can not guarantee such as "permanent package update", "permanent technical support" and other similar transactions after the merchant commitment, please identify the buyer; 3, in the source code at the same time there is a website demonstration and picture demonstration, and the site is inconsistent with the diagram, the default according to the diagram as the dispute evaluation basis (except for special statements or agreement); 4, in the absence of "no legitimate basis for refund", the commodity written "once sold, no support for refund" and other similar statements, shall be deemed invalid; 5, before the shooting, the transaction content agreed by the two parties on QQ can also be the basis for dispute judgment (agreement and description of the conflict, the agreement shall prevail); 6, because the chat record can be used as the basis for dispute judgment, so when the two sides contact, only communicate with the other party on the QQ and mobile phone number left on the systemhere, in case the other party does not recognize self-commitment. 7, although the probability of disputes is very small, but be sure to retain such important information as chat records, mobile phone messages, etc., in case of disputes, it is convenient for seven PAWS to intervene in rapid processing.

View details

Systemhere declaration

1. As a third-party intermediary platform, Qichou protects the security of the transaction and the rights and interests of both buyers and sellers according to the transaction contract (commodity description, content agreed before the transaction); 2, non-platform online trading projects, any consequences have nothing to do with mutual site; No matter the seller for any reason to require offline transactions, please contact the management report.

View details