eGitty

Discover The Most Popular Algorithms

CAER-TABSA: Creating Target and Aspect Representation Based on Context for Targeted Aspect-based Sentiment Analysis Using PyTorch

CAER-TABSA is the pytorch implementation of the paper: Context-aware Embedding for Targeted Aspect-based Sentiment Analysis. It aims to create target and aspect representation based on sentence context for targeted aspect-based sentiment analysist.

Why should we create target and aspect representation?

Targets and aspects may contain same words, which results in targets or aspects having the same vector representations in different contexts and losing the context dependent information.

The structure of CAER-TABSA

CAER-TABSA looks like:

The structure of CAER-TABSA

Datases used in CAER-TABSA

Two benchmark datasets are used to evaluate the performance of CAER-TABSA.

They are: SentiHood, Task 12 of Semeval 2015

Comparative Results of CAER-TABSA with other methods

We can input the target and aspect representation to existing methods to evaluate its effect.

Comparative Results of CAER-TABSA with other methods

From the result we can find: CAER-TABSA can improve the performance of existing models.

Visualization of aspect representation

Here is the visualization of aspect representation

Visualization of aspect representation

How to use CAER-TABSA?

You can run train.py.

You can download source code in here to run: CAER-TABSA Download

Leave a Reply

Your email address will not be published. Required fields are marked *