Abstractive Summarization using AMR and Random Walk Algorithm

The information on Internet is increasing massively in amount. From such enormous availability of data, fetching meaningful knowledge in short span of time is too taxing. With this information deluge, need of automatic summarization has increased twofold. Text summarization is an approach of condensing longer texts without altering it’s meaning and is classified into Extractive and Abstractive Summarization. Former technique reformulates text by extracting related information and concatenating into final summary similar to copy-paste operations. More emphasis is given on sentence structure that is not sufficient to capture gist of information required. The latter makes use of novel words and connectors in forming summary from the aggregated information. The methods of abstractive summarization are less pertinent and tend to ignore semantics. One of the existing techniques, which have graphical approach towards summarization, is Abstract meaning representation (AMR), which is Standard Knowledge framework but ignores identical sentences. The proposed method will build a rooted labeled graph of sentences, checks the similarity of concepts and improves the existing system resulting in fluent summaries.