The Design And Implementation Of Sentiment Analysis Module On HANA System

Nowadays, online social activities have become part of people’s daily lives. Many social media analysis tools appear to provide basic services such as interest words. A couple of companies begin to notice this area, looking forward to an easy, reliable business-oriented social media analysis tool to help the enterprise to do the analysis.To meet these needs, a business software company is trying to build a new social media analysis system on HANA, which provides two key functions, keyword extraction and sentiment analysis, supporting Chinese and English at first. On this system, developers can do either keyword extraction or sentiment analysis quickly and easily by calling the library functions.This paper comes from real practice, using machine learning techniques to design and implement Sentiment Analysis module. Social Media Library, which is the core of social media analysis system, is written in C++ based on LST, Application Framework, SVM, iTab, words segmentation, providing a dynamic library for HANA. Sentiment Analysis module is one of the two modules in Social Media Library, extracting word frequencies and setiment weights as features, using SVM twice to train and predict the sentiments of opinion text into positive, negative or neutral. The training emotional tags are marked artificially. The trained model and classification results are stored in HANA in-memory database. For big data, we use parallel computing techniques for optimization. Basic system dictionaries are combinations of several public online emotion dictionaries.This paper illustrates the background of the system, requirements analysis and takes deep into the design and implementation of Sentiment Analysis module, discribing the whole sentiment analysis processing, especially the feature extraction, classification into three categories, and optimization for SVM parameters. With this system, basic enterprise requirements for social media analysis are satisfied.