A Graph-Based Keyphrase Extraction Model with Three-Way Decision

In recent years, the exploitation of sentence embedding techniques in natural language processing field has encouraged the proposal of new methods for extracting keyphrases from documents based on these techniques. Most of these approaches select keyphrases from a set of candidate phrases based on their semantic proximity to the document. In general, most documents contain complementary paragraphs that are unrelated to the topics covered. This factor reduces the credibility of the semantic proximity of candidate keyphrases to the document. Exploitation of document paragraphs weights during the semantic similarity calculation will inevitably improve the performance of keyphrase extraction from document. In this paper, we propose a new method to extract keyphrases based on document paragraphs weights. Our method is based on sentence embedding techniques and semantic proximity of candidate key phrases from document paragraphs. We evaluated the proposed method on three datasets, Inspec, Semeval2010 and KPTimes, where our results showed that that using document paragraph weight improved the performance of keyphrases extraction.

Unsupervised-learning-based keyphrase extraction from a single document by the effective combination of the graph-based model and the modified C-value method

Keyphrases are multipurpose knowledge gems, rendering keyphrase extraction a very important document processing task. They constitute a concise summary of documents that is extremely useful both for human inspection and machine consumption, in support of a number of tasks in the field of Natural Language Processing, Machine Learning and Information Retrieval. To this regard, the aim of this tutorial is two-fold: (a) to provide an overview of the automatic keyphrase extraction task and (b) to familiarize participants with the keyphrase extraction process. Specifically, we will provide a well‐structured review of the existing work, offer interesting insights on the different evaluation approaches, highlight open issues such as the need for evaluation approaches that take the semantic similarity of predicted and golden keyphrases, present a comparative experimental study of popular techniques, and familiarize the audience with the keyphrase extraction process via a demo presentation using jupyter notebook, which will be available to the audience during the practical part of the tutorial. We expect the tutorial to help newcomers and veterans alike navigate the large amount of prior art and grasp its evolution.

The study of aspect-level sentiment analysis using deep learning methods is one of the more important research directions in the field of natural language processing in recent years. In this paper, we address the problem of insufficient extraction of deep semantic features in existing aspect-level sentiment analysis research, design and build a sentiment analysis model based on the pre-trained language model BERT, fuse BiLSTM and GCN deep learning methods, analyze the sentiment tendency on the collected product review dataset, design relevant experiments to compare in the same application scenario, and verify the effectiveness of the proposed model.

In this paper, we propose a graph-based method for 3D vessel tree structure segmentation based on a new tubularity Markov tree model (TMT), which works as both new energy function and graph construction method. With the help of power-watershed implementation [7], a global optimal segmentation can be obtained with low computational cost. Different with other graph-based vessel segmentation methods, the proposed method does not depend on any skeleton and ROI extraction method. The classical issues of the graph-based methods, such as shrinking bias and sensitivity to seed point location, can be solved with the proposed method thanks to vessel data fidelity obtained with TMT. The proposed method is compared with some classical graph-based image segmentation methods and two up-to-date 3D vessel segmentation methods, and is demonstrated to be more accurate than these methods for 3D vessel tree segmentation. Although the segmentation is done without ROI extraction, the computational cost for the proposed method is low (within 20 seconds for 256{*}256{*}144 image).

A new graph-based semi-supervised method for surface defect classification

Image question answering is becoming a very attractive topic in the field of computer vision and natural language processing in recent years. In this work, we propose a novel Semantic Bi-Embedded Gated Recurrent Unit (SBE-GRU) method to answer fill-in-the-blank style multiple choice questions for images. Different from the single network, we use the SBE-GRU model to learn the high-level semantic information existing in images. To learn the semantic mapping of an image from the visual level to the language level, we feed the visual-language mappings into a stacked GRU. Moreover, to choose the right answer in the candidate options more simply and effectively, we regard the answer sentence as an answer list while training. In the extensive experiments, the proposed method can get better results compared with the state-of-the-art and CCA methods.

In this paper, we present a novel graph-based method for segmenting the whole 3D vessel tree structures. Our method exploits a new adaptive cylinder flux (ACF) based connectivity framework, which is formulated based on random walks. To avoid the shrinking problem of elongated structure, all existing graph-based energy optimization methods for vessel segmentation rely on skeleton or ROI extraction. As a result, the performance of these vessel segmentation methods then depends heavily on the skeleton extraction results. In this paper, with the help of ACF based connectivity framework, a global optimal segmentation result can be obtained without extracting skeleton or ROI. The classical issues of the graph-based methods, such as shrinking bias and sensitivity to seed point location, can be solved effectively with the proposed method thanks to the connectivity framework.

Graph-based method is one of the most efficient unsupervised ways to extract keyword from a single web text. However, rarely did the previous graph-based methods consider the sentence importance. In this paper, we propose a graph-based keyword extractor WS-Rank which brings sentences into graph where sentences are distinctively treated according to their importance. The candidate keywords are extracted through the voting mechanism between words and sentences. To evaluate the experiment, we compare our method with TextRank, a graph-based method which uses the logic distribution relationship only between words. Experiment on 13702 web texts carried out shows that WS-Rank achieves more ideal results with an average F-score of 25.20 %.

Keyphrase extraction is to select the most representative phrases within a given text. While supervised methods require a large amount of training data, unsupervised methods can perform without prior knowledge such as language. In this paper, we propose a ranking algorithm based on unsupervised keyphrase extraction and develop a framework for retrieving opinion articles. Since the proposed algorithm uses an unsupervised method, it can be employed to multi-language systems. Moreover, our proposed ranking algorithm measures the importance in three aspects, the amount of information within articles, representativeness of sentences, and frequency of words. Our framework shows better performance than previous algorithms in terms of precision and NDCG.KeywordsOpinion articleRanking algorithmUnsupervised keyphrase extraction

Keyphrase is an important way to quickly get the topic of a document by providing highly-summative information. The previous approaches for keyphrase extraction simply rank keyphrases according to statistics-based model or graph-based model, which ignore the influence of external knowledge. In this paper, we take prior knowledge, which contains controlled vocabulary of keyphrases and their prior probability, into consideration to enhance previous methods. First, we build a controlled vocabulary of keyphrases introduced by keyphrases from existing collections and a keyphrase candidate set is filtered from a given document by it. Then, we use prior probability to represent the importance of keyphrases candidate with TF-IDF and TextRank. Finally, a supervised learning algorithm is used to learn optimal weights of these three features. Experiments on four benchmark datasets show the great advantages of prior knowledge on keyphrase extraction. Furthermore, we achieve competitive performance compared with the state-of-the art methods.

Sentiment classification is an important topic in the field of natural language processing (NLP). The mainstream methods have also changed from the initial emotional dictionary to the later machine learning, and now most of them focus on deep learning. In sentiment classification tasks, comments in different fields have many similar expressions and relevant characteristics, which can effectively solve the problem of insufficient training data for current deep learning methods. In order to improve the classification effect of the model, we proposed a sentiment analysis model based on multi-task learning that Distillation Bidirectional Encoder Representations from Transformers combined with multi-scale convolution (MTL-DISTILBERT-MSCNN). It models comments in multiple fields together, and uses the advantages of the multi-task learning framework, the pre-training model DistilBert and the convolutional neural network (CNN) model to obtain the correlation between comments in multiple fields, the global features and local features of the text. We conduct experiments on 14 product review data sets and 2 movie review data sets. Using accuracy, F1-score and training speed as the evaluation criteria of the proposed model. By comparing single-task learning and other multi-task learning methods, the MTL-DISTILBERT-MSCNN model can be improved by 2%-5%, and the training efficiency is very high. In addition, we also adjusted the convolutional neural network of the model to further optimize the training effect of the model.

PurposeAutomatic keyphrase extraction (AKE) is an important task for grasping the main points of the text. In this paper, we aim to combine the benefits of sequence labeling formulation and pretrained language model to propose an automatic keyphrase extraction model for Chinese scientific research.Design/methodology/approachWe regard AKE from Chinese text as a character-level sequence labeling task to avoid segmentation errors of Chinese tokenizer and initialize our model with pretrained language model BERT, which was released by Google in 2018. We collect data from Chinese Science Citation Database and construct a large-scale dataset from medical domain, which contains 100,000 abstracts as training set, 6,000 abstracts as development set and 3,094 abstracts as test set. We use unsupervised keyphrase extraction methods including term frequency (TF), TF-IDF, TextRank and supervised machine learning methods including Conditional Random Field (CRF), Bidirectional Long Short Term Memory Network (BiLSTM), and BiLSTM-CRF as baselines. Experiments are designed to compare word-level and character-level sequence labeling approaches on supervised machine learning models and BERT-based models.FindingsCompared with character-level BiLSTM-CRF, the best baseline model with F1 score of 50.16%, our character-level sequence labeling model based on BERT obtains F1 score of 59.80%, getting 9.64% absolute improvement.Research limitationsWe just consider automatic keyphrase extraction task rather than keyphrase generation task, so only keyphrases that are occurred in the given text can be extracted. In addition, our proposed dataset is not suitable for dealing with nested keyphrases.Practical implicationsWe make our character-level IOB format dataset of Chinese Automatic Keyphrase Extraction from scientific Chinese medical abstracts (CAKE) publicly available for the benefits of research community, which is available at:https://github.com/possible1402/Dataset-For-Chinese-Medical-Keyphrase-Extraction.Originality/valueBy designing comparative experiments, our study demonstrates that character-level formulation is more suitable for Chinese automatic keyphrase extraction task under the general trend of pretrained language models. And our proposed dataset provides a unified method for model evaluation and can promote the development of Chinese automatic keyphrase extraction to some extent.

Automatic key phrase extraction is fundamental to the success of many recent digital library applications and semantic information retrieval techniques and a difficult and essential problem in Vietnamese natural language processing (NLP). In this work, we propose a novel method for key phrase extracting of Vietnamese text that exploits the Vietnamese Wikipedia as an ontology and exploits specific characteristics of the Vietnamese language for the key phrase selection stage. We also explore NLP techniques that we propose for the analysis of Vietnamese texts, focusing on the advanced candidate phrases recognition phase as well as part-of-speech (POS) tagging. Finally, we review the results of several experiments that have examined the impacts of strategies chosen for Vietnamese key phrase extracting.

Coreference resolution is an important task in the field of natural language processing. Most existing methods usually utilize word-level representations, ignoring massive information from the texts. To address this issue, we investigate how to improve Chinese coreference resolution by using span-level semantic representations. Specifically, we propose a model which acquires word and character representations through pre-trained Skip-Gram embeddings and pre-trained BERT, then explicitly leverages span-level information by performing bidirectional LSTMs among above representations. Experiments on CoNLL-2012 shared task have demonstrated that the proposed model achieves 62.95% F1-score, outperforming our baseline methods.