---
library_name: paddlenlp
license: apache-2.0
language:
- zh
---
[![paddlenlp-banner](https://user-images.githubusercontent.com/1371212/175816733-8ec25eb0-9af3-4380-9218-27c154518258.png)](https://github.com/PaddlePaddle/PaddleNLP)

# PaddlePaddle/uie-senta-nano

Sentiment analysis is a research hotspot in recent years, aiming at analyzing, processing, summarizing and reasoning emotionally subjective texts. Sentiment analysis has a wide range of application scenarios and can be applied to consumer decision-making, public opinion analysis, personalized recommendation and so on.

According to the analysis granularity, it can be roughly divided into three categories: document-level sentiment analysis, sentence-level sentiment analysis and aspect-level sentiment analysis. Among them, aspect-level sentiment analysis includes multiple subtasks, such as aspect term extraction, opinion term extraction, aspect-opinion-sentiment triplet extraction, etc.


UIE-Senta is a type of Chinese sentiment analysis model, which uses UIE as backbone and further trained based on  large amount of samples related to sentiment analysis. So it has a stronger ability to understand sentiment knowledge and handle the related samples.  Currently, UIE-Senta supports most of basic sentiment analysis capabilities,  including sentiment-level sentiment classification, aspect-term extraction, opinion-term extraction, aspect-sentiment pair extraction, aspect-opinion pair extraction,  aspect-opinion-sentiment triple extraction. You could perform sentiment analysis with UIE-Senta to improve your business analysis capabilities.


<div align="center">
    <img src="https://user-images.githubusercontent.com/35913314/199965793-f0933baa-5b82-47da-9271-ba36642119f8.png" />
</div>


## Available Models

|     Model Name     |           Model Config           |
|  :---------------: |  :-----------------------------: | 
| `uie-senta-base`   | 12-layers, 768-hidden, 12-heads  | 
| `uie-senta-medium` | 6-layers,  768-hidden, 12-heads  | 
| `uie-senta-mini`   | 6-layers,  384-hidden, 12-heads  | 
| `uie-senta-micro`  | 4-layers,  384-hidden, 12-heads  |
| `uie-senta-nano`   | 4-layers,  312-hidden, 12-heads  | 


## Performance on Text Dataset

We conducted experiments to compare the performance different Models based on a self-built test set, which containing samples from multiple fields, such as hotel, restaurant，clothes and so. The comparison results are as follows.

|      Model Name     |  Precision  |    Recall   |      F1     |
|  :----------------: |  :--------: |  :--------: |  :--------: |
|  `uie-senta-base`   |   0.93403   |  0.92795    |  0.93098    |  
| `uie-senta-medium`  |   0.93146   |  0.92137    |  0.92639    |
| `uie-senta-mini`    |   0.91799   |  0.92028    |  0.91913    |
| `uie-senta-micro`   |   0.91542   |  0.90957    |  0.91248    |
| `uie-senta-nano`    |   0.90817   |  0.90878    |  0.90847    |


> Detailed Info: https://github.com/1649759610/PaddleNLP/tree/develop/applications/sentiment_analysis/unified_sentiment_extraction