CityU Seminar on The First Step for AI-based Human-Like Language Understanding – Sentiment Analysis of Text

CityU SEEM Department invited guest speakers from Institute of High Performance Computing (IHPC), Agency for Science, Technology and Research (A*STAR), Singapore to give a seminar named “The First Step for AI-based Human-Like Language Understanding - Sentiment Analysis of Text” on 22^nd Jan 2019.  I attended the seminar and summarized it for sharing. In the beginning, Prof. Min XIE (Chair Professor of Industrial Engineering) introduced guest speakers to us.

Firstly, Dr. Seng Beng Ho (Senior Scientist & Deputy Director, Social and Cognitive Computing Department; Senior Scientist, AI Program, IHPC, A*STAR) introduced Singapore’s Research Ecosystem. 

Then he briefed about A*STAR that had more than 5400 staff in which more than 4500 Researchers, Engineers and Technical Support Staff in different Councils and Research Units.

And then he overviewed the Institute of High Performance Computing, A*STAR and they would be empowering industries on computational simulation, modelling, visualization and AI.  

Dr. Zhaoxia Wang (Programme Manager and Scientist, Social and Cognitive Computing Department, IHPC, A*STAR) was our guest speaker and she shared her research result on AI based Human-Like Language Understanding on Sentiment Analysis of Text.  She would like to develop social media text analysis method to analyse the opinions such as sentiment and emotions in text.  

Then Dr. Wang introduced the Sentiment Analysis Methods that had three methods and they were Machine Learning-based, Non-learning based and Hybrid methods.  For Machine Learning-base Methods, there were many existing methods like NN, FNN, NB, SVM, etc. She selected some of them for enhancement.

And then Dr. Wang proposed enhancement strategies to handle Feature Selection, Negation Dealing and Emoticon Handling.  She evaluated the performance by four metrics such as Precision, Recall, F-measure and Accuracy for employing social media web data sets. 

The following evaluation results for four selected Machine Learning Algorithms using three data sets to compare with and without enhancement methods.  It was found that all enhancement methods had better accuracy than without.

After that Dr. Wang briefed a new methods named Hybrid Methods – Fine-grained Sentiment Analysis.  Firstly, she mentioned the existing problem in the current methods.  

Secondly, she introduced two methodologies named SentiMo (English-based Sensing Method) and ChiEFS (Chinese-based Multi-lingual Hybrid Method).  She would like to achieve human like understanding – mimic human understanding process which realized Fine-grained emotion understanding, Object specificity and Mixed language processing.  

Finally, she demonstrated different translation results using Google, Baidu, Microsoft translator to test their accuracy.  Lastly, she briefed her Hybrid Multilingual Processing Unit with Submodule Design and she had been applying patent filing for Intelligent Sensing with Adaptive Learning model.  

After the seminar, we had lunch together.

Reference:

SEEM Dept., CityU - http://www.cityu.edu.hk/seem/

The First Step for AI-based Human-Like Language Understanding - Sentiment Analysis of Text - http://www.cityu.edu.hk/seem/pdf/seemseminar1819_003.pdf