Similarity-Based Estimation for Document Summarization using Fuzzy Sets

Information is increasing every day and thousands of documents are produced and made available in the Internet. The amount of information available in documents exceeds our capacity to read them. We need access to the right information without having to go through the whole document. Therefore, documents need to be compressed and produce an overview so that these documents can be utilized effectively. Thus, we propose a similarity model with topic similarity using fuzzy sets and probability theories to extract the most representative sentences. Sentences with high weights are extracted to form a summary. On average, our model (known as MySum) produces summaries that are 60% similar to the manually created summaries, while tf.isf algorithm produces summaries that are 30% similar. Two human summarizers, named P1 and P2, produce summaries that are 70% similar to each other using similar sets of documents obtained from TREC.