Improved Speech Recognition, Classification, Extraction using Conditional Random Field with Kernel Approach

摘要

Extracting useful information from the pool of big data gives birth to new domain known as Information Extraction. The domain of Information Extraction has its genesis in Natural Language Processing (NLP). The fundamental drift in this field takes the birth from various competitions that are focused on the recognition and extraction of named entities such as names of people, organizations etc. As the world become more data oriented by advent of Internet, new applications of processing of structured and unstructured data comes in light. Most of the interest is to extract and classify named entities like person, organization and location etc. that is a subtask of Information Extraction known as Entity Extraction and Classification. In this field, number of models, from handcrafted rules to unsupervised learning techniques, were proposed but extracting entities and then classifying them with inference of data present before and after it with great accuracy is still a bottleneck. In this thesis work, a model is proposed that is using Kernel function for reducing the overlapping of information and Conditional Random Field (CRF) for predicting dependency among features. The results will be analyzed by using parameters like precision, recall and accuracy.