Using Statistical and Semantic Models for Multi-Document Summarization

The 2018 Conference on Computational Linguistics and Speech Processing ROCLING 2018, pp. 169-183 ©The Association for Computational Linguistics and Chinese Language Processing Using Statistical and Semantic Models for Multi-Document Summarization Divyanshu Daiya Department of Computer Science Engineering LNM Institute of Information Technology Jaipur, INDIA daiyadivyanshu@gmail.com Anukarsh Singh Department of Computer Science Engineering LNM Institute of Information Technology Jaipur, INDIA anukarshsingh1@gmail.com Abstract. We report a series of experiments with different semantic models on top of various statistical models for extractive text summarization. Though statistical models may better cap- ture word co-occurrences and distribution around the text, they fail to detect the context and the sense of sentences /words as a whole. Semantic models help us gain better insight into the context of sentences. We show that how tuning weights between different models can help us achieve significant results on various benchmarks. Learning pre-trained vectors used in seman- tic models further, on given corpus, can give addition spike in performance. Using weighing techniques in between various statistical models too further refines our result. For Statistical models, we have used TF/IDF, TextRAnk, Jaccard/Cosine Similarities. For Semantic Models, we have used WordNet-based Model and proposed two models based on Glove Vectors and Facebook’s InferSent. We tested our approach on DUC 2004 dataset, generating 100-word summaries. We have discussed the system, algorithms, analysis and also proposed and tested possible improvements. ROUGE scores were used to compare to other summarizers. Keywords. Extractive Text Summarization, Semantic Summarization Models, Statistical Summarization Models, Multi Document Summarization 169 1. Introduction Automatic Text Summarization deals with the task of condensing documents into a summary, whose level is similar to a human-generated summary. It is mostly distributed into two distinct domains, i.e., Abstractive Summarization and Extractive Summarization. Abstractive sum- marization( Dejong et al. ,1978) involves models to deduce the crux of the document. It then presents a summary consisting of words and phrases that were not there in the actual document, sometimes even paraphrasing [20]. A state of art method proposed by Wenyuan Zeng [25] pro- duces such summaries with length restricted to 75. There have been many recent developments that produce optimal results, but it is still in a developing phase. It highly relies on natural lan- guage processing techniques, which is still evolving to match human standards. These short- comings make abstractive summarization highly domain selective. As a result, their application is skewed to the areas where NLP techniques have been superlative. Extractive Summarization, on the other hand, uses different methods to identify the most informative/dominant sentences through the text, and then present the results, ranking them accordingly. In this paper, we have proposed two novel stand-alone summarization methods.The first method is based on Glove Model [17],and other is based on Facebook’s InferSent [4]. We have also discussed how we can effectively subdue shortcomings of one model by using it in coalition with models which capture the view that other faintly held. 2. Related Work A vast number of methods have been used for document summarization. Some of the methods …

Content truncated to 500 words