Entity Extraction for Malayalam Social Media Text Using Structured Skip-gram Based Embedding Features from Unlabeled Data

Social media text is generally informal and noisy but sometimes tends to have informative content. Extracting these informative content such as entities is a challenging task. The main aim of this paper is to extract entities from Malayalam social media text efficiently. The social media corpus used in our system is from FIRE2015 entity extraction task. This data is initially subjected to pre-processing and feature extraction and then proceeds with entity extraction. Apart from the conventional stylometric features like prefixes, suffixes, hash tags etc., and POS tags, unsupervised word embedding features obtained from Structured Skip-gram model are utilized to train the system. The extracted features is given to the Support vector machine classifier to build and train model. Testing of the system resulted in better accuracy than the existing systems evaluated in FIRE2015 tasks. Unsupervised features retrieved using Structured Skip-gram model contributes to the reason for achieving better performance.