Textstar: a Fast and Lightweight Graph-Based Algorithm for Extractive Summarization and Keyphrase Extraction

Textstar: a Fast and Lightweight Graph-Based Algorithm for Extractive Summarization and Keyphrase Extraction David Brock Dallas College dbrock@dallascollege.edu Ali Khan University of North Texas alikhan@my.unt.edu Tam Doan University of North Texas tamdoan@my.unt.edu Alicia Lin University of North Texas alicia.y.lin@gmail.com Yifan Guo University of North Texas yifan.guo.3517@gmail.com Paul Tarau University of North Texas paul.tarau@unt.edu Abstract We introduce Textstar, a graph-based summarization and keyphrase extraction system that builds a document graph using only lemmatization and POS tagging. The document graph aggregates connections between lemma and sentence identifier nodes. Consecutive lemmas in each sentence, as well as consecutive sentences themselves, are connected in rings to form a "ring of rings" representing the document. We iteratively apply a centrality algorithm of our choice to the document graph and trim the lowest ranked nodes at each step. After the desired number of remaining sentences and lemmas is reached, we extract the sentences as the summary, and the remaining lemmas are aggregated into keyphrases using their context. Our algorithm is efficient enough to process large document graphs without any training, and empirical evaluation on several benchmarks indicates that our performance is higher than most other graph-based algorithms. 1 Introduction Contemporary natural language processing is mostly done through neural networks. However, this is resource intensive and requires large amounts of data. This can be a problem for languages that are not widely spoken, due to insufficient data for training these models. Even for tasks where neural network based models excel, they are often an overkill. State of the art Transformer-based tools such as BERT (Devlin et al., 2019), GPT-3 (Brown et al., 2020), and even Longformer (Beltagy et al., 2020) have size limits and require hierarchical approaches to long documents. However, graph-based approaches are one-shot algorithms that do not require expensive computational resources to train. The need for a fast graph-based summarizer is also justified as a preprocessor to assist these neural models by enabling them to work on salient smaller subsets of a large document. To address this insufficiency, we propose Textstar, a lightweight graph-based summarization and keyphrase extraction algorithm that outperforms most other graph-based methods. Our model is language-independent, making it suitable for application to languages with insufficient training data. Additionally, it simultaneously supports keyphrase extraction and summarization. This flexibility, along with its short runtime, opens up possibilities for many applications. We will start with an overview of our system, as presented in Figure 1. After a document is uploaded, it is pre-processed. Then, each sentence is converted into a connected ring of nodes (Figure 2). Additionally, the sentence IDs are also connected into a ring of nodes. Afterwards, the graph is fed to the Textstar algorithm, which gradually trims out word nodes and sentence ID nodes that have low ranking values. When the graph contains only the desired number of sentence ID nodes or keyword nodes, we feed this information to the postprocessing component. The postprocessing component then converts certain keywords into keyphrases and combines the sentences represented by the sentence ID nodes into …

Content truncated to 500 words