Performance of Content Based Mining Approach for Multi - lingual Textual Data

Data mining has become a necessary and powerful tool in the present era of web and internet communications. It has also evolved into media mining wherein heterogeneous data inputs like figures, videos and audios are gradually getting embedded into the web and this makes it quite complex and different. These and other aspects like currency and ‘liveliness’ of the web bring in more interesting features making a shift from translation to especially content extraction. Content extraction in web pages with Indian regional languages or English as the parent language have many aspects like free use of one language in another like ‘computer’ being used as it is with regional text and inclusion of other forms of data like hand written texts or sketches or drawings. This is common in education, news and entertainment and the focus of the current paper is in extracting content in a hybrid document with hand-written texts embedded. Work has been carried out initially with web documents in the form of computer generated text since they are more crisp in nature. Extending the idea, the present paper discusses on the results of hand-written text format and a comparative study with computer generated text format, which are less crisp in nature and more fuzzy depending on the writer. Beginning with letters having common content to words with common content, results of features on pixel maps are presented first. Later extraction using normalisation studies and classification means are presented.