Unsupervised Summarization Approach With Computational Statistics of Microblog Data

Introduction

In recent time, Twitter is recognised as one of the well accepted Microblogging platform which started in 2006. More than 340 million Tweets are daily posted by more than 500 million of users. The majority of the posts short messages as microblog accept limited text which is known as Tweets which are informal or not particularly significant, about 3.6% (Wikipedia, 2021) of the Tweets represent topics of conventional news. During the time of any important event an enormous set of posts are made between a large numbers of users. This huge rate of message exchange causes information overload. Twitter provides a list of popular topics which are known as Trending Topics in chronological order. For researchers, Twitter makes available an API to download data by streaming approach in chronological order that contains a topic (hashtag)(REST API Resources, 2021)(Sandy Hook Elementary School shooting, 2012). All the data are sorted by recency irrespective of relevancy. If anybody has to understand any topic, the user has to go through all the tweets which are almost a tedious job. This is one of the challenging areas in microblog analysis. So, several researchers are currently focusing on Tweet summarization to handle this information retrieval in a more efficient way.

Huge range of users is posting millions of messages in social media during any important events. And not only valid users, spammers are also using the same platforms to spared spam content and fake news as well. There are two different types of summarization, abstractive and extractive. Abstractive summarization is the process of reducing large text content into short text/paragraphs that convey a concise meaning of the original content. The large content needs to be made concise to make out meaning from the huge amount of data. Extractive summarization is the process to identify selected portion text which can represent the entire database. In several text summarization research works a large number of researches are carried out, and mostly on microblog dataset. Microblog data is redundant and irrelevant in nature, therefore, to summarize any huge dataset redundant and similar types of data need to be categorized. So, clustering on this social media that is another correlated challenge comes with summary generation. Similar data are posted multiple times so categorization is another challenging task by grouping syntactical and semantic similar content together. The key issue of microblog summarization is characteristically a multi-document summarization problem. Nonetheless, the cutting edge techniques in summarization of single document are additionally important, by utilizing the microblogs as inputs to make up a single document. It has some discrete constraints, fundamentally because of the tiny size of particular microblogs, and the casual representation of texts in microblogs, that turns it complex to decipher the semantic similarities of microblogs. The earlier state of the art research shows that various researches had been carries out such as – (1) Cluster-rank, (2) COWTS, (3) Frequency Summarizer, (4) LexRank, (5) LSA, (6) LUHN, (7) Mead, (8) SumBasic and (9) SumDSDR. These algorithms mostly select dissimilar sets of tweets in the output summaries. Therefore, different summarization methods typically choose diverse groups of tweets in the summaries for the similar set of input tweets. The distinctive summarization techniques are probably going to assess the general significance of tweets dependent on various components, and thus different rundowns are probably going to reveal features of the input dataset.