SNS 특징정보를 활용한 단문텍스트의 카테고리 분류에 관한 연구

Abstract

Recently, developments in wireless internet and the spread of mobile devices, social networking services (SNS) are rapidly progressing. SNS allows the user to place one's offline social connections onto an online platform, which in turn, strengthens and widens the user's personal network. As the number of SNS users rapidly increases, SNS is being transformed into a platform that allows the user to share one's interests with an unspecified mass, which, in the process, is changing the objective of SNSs from strengthening user's personal networks into an informational platform that freely shares interests or information. Due to some characteristics of SNS, including the real-time generation and rapid spread of data, there has been a phenomenon of decreased efficiency in information acquisition using search engines, largely because of the overload of available information. To alleviate the problem of information overload, there have been experiments that explore the realm of document sorting, search engines, text translation, spam mail filtering, and many others. Data that generated on the internet or by social networking do not have a consistent standard, and because the contents of said data include short sentences, it is hard to sort the data based just on the content of the document. Therefore, this dissertation attempted to solve the phenomenon of overload of information due to the large amount of data generated real time, and to separate the texts used on the internet and other SNSs for effective data management. Short texts were collected from Twitter, which most often displays sentences in short formats. The collected short texts were analyzed to extract certain features. Words were weighted based on the extracted features to examine the correlation between the collected dataset and unsorted data, and the short texts were categorized. From the results of the experiment proposed in this dissertation, it was found that text categorization showed better performance when features were weighted using feature frequencies. Result of short text classification utilizing the feature frequency , it shows the 90.7 percent .