JASen

KO | EN

gitlite — search

#python #hacktoberfest #llm #rust #deep-learning #docker #java #nodejs #windows #awesome #awesome-list #security #ios #openai

(★ 49)

Code and Data for our EMNLP-2020 paper Weakly-Supervised Aspect-Based Sentiment Analysis via Joint Aspect-Sentiment Topic Embedding.

JASen 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Weakly-Supervised Aspect-Based Sentiment Analysis via Joint Aspect-Sentiment Topic Embedding

The code and data used for our EMNLP [paper](https://arxiv.org/abs/2010.06705) Weakly-Supervised Aspect-Based Sentiment Analysis via Joint Aspect-Sentiment Topic Embedding.

## Requirements

* GCC compiler (used to compile the source c file): See the [guide for installing GCC](https://gcc.gnu.org/wiki/InstallingGCC).

## Datasets

We collect in-domain corpus for embedding training. For evaluation, we use Restaurant and Laptop datasets in [Sem-Eval 2015](http://alt.qcri.org/semeval2015/task12/) and [Sem-Eval 2016](http://alt.qcri.org/semeval2016/task5/). We preprocessed these datasets in this repository.

## Run the Code

### Using the same datasets as ours
```
bash run_jasen.sh
```
This step runs the whole pipeline from embedding training, to neural network distillation and model evaluation. The ``--dataset`` in the script is used to specify which prepared dataset (restaurant or laptop) to use. Generated embedding file is stored under ``${dataset}``.
Prediction results for each dataset are generated at ``/datasets/${dataset}/prediction.txt``.

### Preparing your own dataset
Create a new folder under ``/datasets`` for your new dataset. The in-domain unlabeled training corpus ``train.txt`` used for joint topic embedding training has the format of each line being a document. The test set ``test.txt`` used for evaluation is in following format:
```
line_id	aspect_label_id	sentiment_label_id	text
```
The keywords for each aspect and sentiment should be listed in ``aspect_w_kw.txt`` and ``senti_w_kw.txt``. Each line refers to one aspect/sentiment category. The line order should be consistent with the order of aspect and sentiment label ids. Examples can be found in prepared dataset folders.

JASen

JASen 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

JASen

JASen 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README