Text-based Features Research Articles

Readability of Patient-Reported Outcome Measures Used in Voice Disorders

ObjectivesThis study aims to evaluate Turkish patient-reported outcome measures (PROM) used in voice disorders using different readability formulas and to examine the relationship between text-based features and readability values. Study DesignA search was conducted to identify published Turkish PRO questionnaires related to dysphonia that are completed by patients. MethodsSearch was conducted to identify published PROM related to dysphonia that are completed by patients in a self-administered format. Reading grade levels were analyzed for voice-related PROM using the Ateşman readability and the Bezirci-Yılmaz new readability formulas as computed by a readability calculations software package. Text-based quantitative features are obtained. Descriptive statistics were also computed across the questionnaires. ResultsA total of 14 PROMs were found. In the analysis, the mean of the Ateşman readability values of the PROM was calculated as 70.66 (±5.55). This value is defined as 'easy', according to the Ateşman readability measures. The average Bezirci-Yılmaz new readability value is 6,03 (±1,07). This value corresponds to the 6th grade reading level in the Turkish education system. ConclusionsResults show that more than half of the PROMs in the Turkish language have the recommended reading level. PROMs allow healthcare professionals to objectively understand the patient's symptoms, both qualitatively and quantitatively. Therefore, it is critical that patients accurately understand and respond to these outcome measures. This is why developers should consider reading as a testable structure. Speech and language therapists should consider the reading level required to understand a specific PROMs in order to obtain the most useful information for treatment planning and outcome assessments.

Stance Detection in the Context of Fake News—A New Approach

Online social networks (OSNs) are inundated with an enormous daily influx of news shared by users worldwide. Information can originate from any OSN user and quickly spread, making the task of fact-checking news both time-consuming and resource-intensive. To address this challenge, researchers are exploring machine learning techniques to automate fake news detection. This paper specifically focuses on detecting the stance of content producers—whether they support or oppose the subject of the content. Our study aims to develop and evaluate advanced text-mining models that leverage pre-trained language models enhanced with meta features derived from headlines and article bodies. We sought to determine whether incorporating the cosine distance feature could improve model prediction accuracy. After analyzing and assessing several previous competition entries, we identified three key tasks for achieving high accuracy: (1) a multi-stage approach that integrates classical and neural network classifiers, (2) the extraction of additional text-based meta features from headline and article body columns, and (3) the utilization of recent pre-trained embeddings and transformer models.

Examining the merits of feature-specific similarity functions in the news domain using human judgments

Online news article recommendations are typically of the ‘more like this’ type, generated by similarity functions. Across three studies, we examined the representativeness of different similarity functions for news item retrieval, by comparing them to human judgments of similarity. In Study 1 (N=401\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$N=401$$\\end{document}), participants assessed the overall similarity of ten randomly paired news articles on politics and compared their judgments to different feature-specific similarity functions (e.g., based on body text or images). In Study 2, we checked for domain differences in a mixed-methods survey (N=45\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$N=45$$\\end{document}), surfacing evidence that the effectiveness of similarity functions differs across different news categories (‘Recent Events’, ‘Sport’). In Study 3 (N=173\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$N=173$$\\end{document}), we improved the design of Study 1, by controlling for how news articles were matched, differentiating between dissimilar news articles and articles that were matched on a shared topic, named entities, and/or date of publication, across ‘Recent Events’ and ‘Sport’ categories. Across all studies, we found that users mostly used text-based features (e.g., body text, title) for their similarity judgments, while BodyText:TF-IDF was found to be the most representative for their judgments. Moreover, the strength of similarity judgments by humans and similarity scores by feature-specific functions was strongly affected by how news article pairs were matched. We show that humans and similarity functions are better aligned when two news articles are more alike, such as in a news recommendation scenario.

ICA-CRMAS: Intelligent Context-Awareness Approach for Citation Recommendation based on Multi-Agent System

Navigating the ever-expanding sea of scientific literature presents a daunting challenge for researchers seeking relevant and up-to-date information. Traditional citation recommendation systems, while well-intentioned, often fall short due to their limited focus on text-based features and lack of contextual awareness. In this paper we introduce the ICA-CRMAS (Intelligent Context-Aware Approach for Citation Recommendation based on Multi-Agent System), an intelligent system that leverages the power of deep learning, semantic analysis, and multimodal learning to overcome these limitations. ICA-CRMAS goes beyond the surface, delving into the rich tapestry of information within academic papers, including figures, which often hold vital contextual clues. By weaving this contextual data directly into its recommendation models, ICA-CRMAS generates highly personalized and relevant suggestions. This comprehensive approach unlocks enhanced accuracy, diversity, and serendipity, enabling researchers to effectively discover papers aligning with their interests and research objectives. ICA-CRMAS illuminates its reasoning. Instead of opaque suggestions, the system provides clear explanations that justify and illustrate recommended citations. This transparency builds user confidence, allowing researchers to critically engage with and trust the system’s recommendations. Evaluation experiments conducted on real-world academic datasets demonstrate that ICA-CRMAS outperforms existing approaches across various metrics. it surpassing its closest competitor by a margin of 7.53 on accuracy, 6.07% on MRR and by 5.87 on Recall. User feedback further reinforces its effectiveness, with an Overall System Usability Scale (SUS) score of 76.73, exceeding benchmark scores for comparable systems.

Research on Fake News Detection Based on Dual Evidence Perception

With the development of online social networks, the prevalence of fake news on social media is increasing, leading to a heightened focus on detecting fake news. To combat this, researchers have developed various strategies, including text-based and fact-checking approaches. The text-based approach analyses linguistic features within fake news posts, while the fact-checking method verifies statements against an external knowledge base, disregarding the linguistic patterns between posts. Despite their differences, both approaches offer complementary benefits in the fight against fake news, highlighting the importance of integrating these methods. However, there has been relatively limited investigation into merging these two strategies into a unified framework. In this paper, we proposed the FakE News Detection Based on Dual Evidence Perception (DEP-FEND) method. This novel approach merges text-based and fact-checking techniques into a unified framework to enhance fake news detection. DEP-FEND consists of three sub-modules: evidence perception based on the historical news environment, evidence perception from an external knowledge base, and the fake news detector. Initially, two independent modules were developed to gather evidence from the historical news environment and an external knowledge base using the Gaussian kernel soft-counting method, generating two distinct evidence vectors. Subsequently, these two independent perception modules were combined using domain gates, and the posts, along with the combined evidence perception signals, were inputted into the fake news detector. The experimental results on the two Chinese datasets demonstrate that our proposed solution outperformed other fake news detection methods with an average accuracy improvement of 3.6% and a macro-f1 improvement of 2.1%.

Explainable text-based features in predictive models of crowdfunding campaigns

AbstractReward-Based Crowdfunding offers an opportunity for innovative ventures that would not be supported through traditional financing. A key problem for those seeking funding is understanding which features of a crowdfunding campaign will sway the decisions of a sufficient number of funders. Predictive models of fund-raising campaigns used in combination with Explainable AI methods promise to provide such insights. However, previous work on Explainable AI has largely focused on quantitative structured data. In this study, our aim is to construct explainable models of human decisions based on analysis of natural language text, thus contributing to a fast-growing body of research on the use of Explainable AI for text analytics. We propose a novel method to construct predictions based on text via semantic clustering of sentences, which, compared with traditional methods using individual words and phrases, allows complex meaning contained in the text to be operationalised. Using experimental evaluation, we compare our proposed method to keyword extraction and topic modelling, which have traditionally been used in similar applications. Our results demonstrate that the sentence clustering method produces features with significant predictive power, compared to keyword-based methods and topic models, but which are much easier to interpret for human raters. We furthermore conduct a SHAP analysis of the models incorporating sentence clusters, demonstrating concrete insights into the types of natural language content that influence the outcome of crowdfunding campaigns.

Suç Kategorisi Tespiti için Yığınlama Topluluk Öğrenimi Modeli Kullanan Çatı Tasarımı

Crime refers to an action legally defined as harmful to society, and it is important to understand the type of crime to prevent these actions. However, crime can occur at any time and place, making it difficult to predict. Data generated based on previously committed crimes contributes to overcoming this difficulty. This study proposes a novel model for classifying criminal activities using a Doc2Vec that can cause a numerical representation of texts regardless of length and a stacking ensemble model that includes 8 different machine-learning models. Unlike the literature, the model processes the features as text and converts them into vectors rather than categorically. In this way, it enables using features that cannot be used in the literature. The proposed model is tested using a distributed online competition database, Francisco Crime Classification, which contains crimes committed over 12 years. An accuracy value of 99.28% was obtained for the 15 crime categories with the highest crime records, while precision, recall, and f-score values were 99.18%, 99.38%, and 99.20%, respectively. With cross-validation (k=10), 99.80% performance was achieved with a std. value of 0.001. These performance values are higher than those of all the studies in the literature using categorical feature structures. The results show that converting criminal activity reports, which contain text-based features, into vectors that can be processed with natural language processing techniques such as Doc2vec instead of using them categorically in model training can directly contribute to the classification performance and provide a more efficient model with less preprocessing.

Feature Classification and Extreme Learning Machine Based Detection of Phishing Websites

Phishing is a cyber-attack that uses a phishing website impersonating a real website to deceive internet users into disclosing sensitive information. Attackers using stolen credentials not only utilize them for the targeted website, but they may also be used to access other famous genuine websites. This paper proposes a novel approach for detecting phishing websites using a feature classification technique and an Extreme Learning Machine (ELM) algorithm. The proposed system extracts various features from the website URL and content, including text-based, image-based, and behavior-based features. These features are then classified using a feature selection technique, which selects the most relevant features to improve the detection accuracy. The selected features are then fed into the ELM algorithm, which is a powerful machine learning method for classifying and predicting data. The ELM algorithm It trains upon a huge set of data legitimate & phishing websites, and final outcome model is applied to classify unknown websites as either legitimate or phishing. The proposed approach is evaluated on several benchmark datasets and compared with other state-of-the-art phishing detection methods.
 The experimental results demonstrate that the proposed approach achieves high detection accuracy and outperforms other methods in terms of precision, recall, and F1-score. The proposed approach can be used as an effective tool for detecting and preventing phishing attacks, which are a major threat to the security of online users.

Hybrid approach for text categorization: A case study with Bangla news article

The incredible expansion of online texts due to the Internet has intensified and revived the interest of sorting, managing and categorising the documents into their respective domains. This shows the pressing need for automatic text categorization system to assign a document into its appropriate domain. In this article, the focus is on showcasing the effectiveness of a hybrid approach that works elegantly by combining text-based and graph-based features. The hybrid approach was applied on 14,373 Bangla articles with 57,22,569 tokens collected from various online news corpora covering nine categories. This article also presents the individual application of both the features to explicate how they generally work. For classification purposes, the feature sets were passed through the Bayesian classification methods which yield satisfactory results with 98.73% accuracy for Naïve Bayes Multinomial (NBM). Also, to test the robustness and language independency of the system, the experiments were performed on two popular English datasets as well.

Language and vision based person re-identification for surveillance systems using deep learning with LIP layers

Real-time surveillance systems have become a necessity of today's life owing to their relevance in the contemporary era for security reasons to ensure a secure and safe environment. Presently, Person re-identification (Re-ID)-based surveillance systems are becoming increasingly more prevalent and sophisticated since they do not require human intervention and are more reliable to deploy in public spaces leveraging multi-camera networks. However, one of the major problems in Person ReID is the visual appearance i-e the appearance of a person in an image is greatly affected by different camera views. As a result, the discriminative set of features must be learned in a deep learning model in order to re-identify persons from opposing camera viewpoints. To address this challenge, we propose an image/text-retrieval-based Person ReId method in which both visual and text-based features are exploited to carry out person re-identification. More precisely, the textual descriptions of the images are taken into account as text features with Glove Word Embedding followed by 1D-MAPCNN and fused with image-level features extracted using the GoogLeNet model. In addition, the feature discriminability is enhanced using local importance-based pooling (LIP) layers in which adaptive significance weights are learned during downsampling. Moreover, from two different modalities, feature refinement is done during training with the help of attention mechanisms using the Convolutional Block Attention module (CBAM) and the proposed shared attention neural network. It is observed that LIP layers along with both vision and textual features are playing a key role in acquiring discriminative features even if the visual appearance of the same person is greatly affected due to camera pose conditions. The proposed method is validated on the CUHK-PADES dataset and has 15.34% and 24.39% rank-1 improvement in text and image-based retrievals.

Rhetorical Structure Theory: Towards Writing with a Sense of Purpose

In recent years, different writing researchers have stressed the importance of integrating text-based and reader-based approaches in writing instruction to enable EFL learners to write both coherently and purposefully. However, combining these two basic features in students’ writing remains a challenging task for both students and teachers. Given this challenge, the present paper describes a model of writing instruction that combines text-based and reader-based features based on both Rhetorical Structure Theory and the process genre approach to improve EFL learners’ writing quality.

Using Multi-Label Multi-Class Support Vector Machines with Semantic and Lexical Features for Aspect Category Detection

In contrast to the aspects, aspect categories are often coarser and don't always appear as terms in sentences. Besides, the typical way to element the types associated with part is generally grainier concerning factors and doesn't exist within verdicts. The primary intent of the study is to investigate the efficacy of Lexicon, linguistic, vector-based, and features correlated to semantics within the aspect of the responsibility built with the finding of aspect category detection ACD). Semantic and emotional data are captured via vector-based features. Further, it examines vector-based feature superiority issues within the compression of features of text-based characteristics. Study purposes to the linguistic efficacy with the Lexicon, linguistic, and semantic features, also vector-based dependent to the system. Also, the information led with vector-based features that capture the semantic with sentimental analysis characteristics. With the experimental outcomes, the performance efficacy with the vector-based features outperformed text-based features. The methodologies associated with deep learning have generated features within the vector orientation relevant to the word-based structures. Therefore, the proposed method achieved effectiveness with the determined constraints by applying the metrics of precision, recall, and F1 scores. Correlating with the performance of ABSA's state-of-the-art techniques, the proposed research process gained superior outcomes.

HELPHED: Hybrid Ensemble Learning PHishing Email Detection

Phishing email attack is a dominant cyber-criminal strategy for decades. Despite its longevity, it has evolved during the COVID-19 pandemic, indicating that adversaries exploit critical situations to lure victims. Plenty of detectors have been proposed over the years, which mainly focus on the contents or the textual information of emails; however, to cope with the evolution of phishing emails more sophisticated approaches should be introduced that will exploit all the emails’ traits to enhance the detection capability of Machine Learning/Deep Learning classifiers. To tackle the limitations of existing works, this paper proposes a phishing email detection methodology, named HELPHED that focuses on the detection of phishing emails by combining Ensemble Learning methods with hybrid features. The hybrid features provide an accurate representation of emails by fusing their content and textual traits. We propose two methods of HELPHED, the first one employs the Stacking Ensemble Learning method, while the second method utilizes the Soft Voting Ensemble Learning. Both methods deploy two different Machine Learning algorithms to handle the hybrid features separately, yet in parallel, minimizing the features’ complexity and improving the model’s performance. A thorough evaluation analysis is carried out considering innovative guidelines that aim to prevent partial and misleading results. Experimental tests verified that the combination of hybrid features with Ensemble Learning, overall, accomplishes better detection performance than when employing only content-based or text-based features. Numerical results on a rich imbalanced dataset (i.e., 32,051 benign and 3,460 phishing email samples) that considers the evolution of phishing emails show that Soft Voting Ensemble Learning outperforms other prominent Machine Learning/Deep Learning algorithms and existing works yielding F1-score equal to 0.9942.

Privacy preserved incremental record linkage

Using an incremental approach to solve the record linkage problem is a relatively new research area. In incremental record linkage, every inserted record is compared with some existing clusters of records based on its blocking key value. Then, considering similarity, either the record will be put into an existing cluster, or a new cluster will be created for it. Although few papers have presented their solutions for incremental record linkage targeting the linkage quality or efficiency, privacy issue regarding the approach has not yet been discussed. Privacy is a major concern when record linkage is performed for sensitive data, e.g., health records, financial records, etc. In this regard, we have come up with a novel concept privacy-preserving incremental record linkage (PPiRL) which encapsulates privacy-preserving techniques with an incremental record linkage approach. In this chapter, we have proposed an end-to-end framework as our solution for PPiRL. For preserving privacy, we have used two types of privacy techniques namely phonetic encoding and generalization. We have used a recently developed phonetic algorithm “nameGist” to handle text-based features. For generalization, we have used the K-anonymization algorithm for numeric and categorical features. For handling incremental updates and internal linkage, we have used the Naive incremental clustering approach using Hierarchical Agglomerative clustering as the base clustering algorithm. We have performed various experiments to test the privacy and linkage quality of PPiRL. We have compared our work with the existing incremental record linkage framework and also with existing privacy-preserved record linkage techniques. It is apparent from our results that other than a small trade-off in linkage quality, our framework works better as a combined package of privacy and linkage solutions that any existing frameworks do not yet provide.

Multimodal fusion methods with deep neural networks and meta-information for aggression detection in surveillance

Multimodal fusion has become one of the hottest topics in affective computing and other research areas. Yet, this topic is less studied in surveillance systems. In general, it focused only on a single modality mainly video or audio. Nevertheless, the fusion of multiple data from different sources and modalities is challenging especially for the detection of aggression because of its complexity. Indeed, aggression is difficult to be defined due to its ambiguity. Therefore, this high level concept is hard to be described through simple rules or trivial algorithms and classifiers. This article presents an approach based on four multimodal fusion methods using the audio, video and text modalities, as well as extra-information with Deep Learning techniques to detect aggression in surveillance. We investigate the combination of acoustic, visual and text-based features, as well as a set of five meta-features which have an influence on the fusion process and the aggression intensity. The proposed architectures use Deep Neural Networks (DNNs) although the dataset is relatively small. The neural network models show their effectiveness not only in prediction tasks, but also in feature extraction and dimension reduction. The first multimodal fusion method is based on an intermediate level that contains the predictions of audio, video and the five meta-features by implementing multiple DNNs. As our previous works on aggression detection based on an intermediate level and deep neural networks on a restricted subset of data also indicate the same findings, we extend these works by working on the whole dataset. The second fusion method uses the concatenation of the different features with the meta-features as labels without the intermediate level. The other two fusion methods are based on element-wise operations as a first step then on the concatenation as a second one. One method uses the element-wise product and the other one uses the element-wise addition. Among these proposed models, we find that the second method surpasses the performance of the other methods with an unweighted average accuracy of 85.66%, and weighted average accuracy of 86.35%, as well as the existing similar approach, for the detection of aggression, that provided accuracies of 75% and 77% for the unweighted average and weighted average. Overall, we think that each fusion method has its advantages and shortcomings for the prediction of different aggression intensities. Furthermore, we observe that meta-features are very informative and show significant improvements for the prediction performance of all the fusion methods.

Online social network individual depression detection using a multitask heterogenous modality fusion approach

In recent years, the number of people who endanger their lives has been increasing rapidly due to the mental burden of depression. The online social network (OSN) provides researchers with another perspective for detecting individuals suffering from depression. However, existing machine learning-based depression detection studies still leave relatively low classification performance, suggesting that there is significant improvement potential in their feature engineering. In this paper, we manually build and publish a large dataset on Sina Weibo (a leading OSN with the largest number of active users in the Chinese community), namely the Weibo User Depression Detection Dataset (WU3D). It includes more than 20,000 normal users and more than 10,000 depressed users, both of which are labeled and rechecked following the DSM-5 official medical and psychological depression document by professionals. Then, we conclude and propose ten statistical features by analyzing the user’s text, social behavior, and posted pictures. In the meantime, text-based word features are extracted using the popular pretrained model XLNet. Moreover, we fuse these features from heterogeneous modalities and implement a multitask learning scheme to train our proposed deep neural network classification model, i.e. FusionNet (FN). The experimental results show that FN has excellent to recognize depressed users on the OSN, achieving the highest F1-Score of 0.9772 on the test set. Compared to existing studies, the proposed method has better classification performance and robustness for unbalanced training samples, as well as reasonable training and inference time. Our work provides a method to fuse multimodal information to detect individual-level depression and has reference significance for similar studies on other OSNs.

A Deep Learning Approach for Robust Detection of Bots in Twitter Using Transformers Model

The volume of audio visual content produced on social networks has increased tremendously in recent decades, and this information is quickly spread and consumed by a large number of people. The disruption of false news sources and bot accounts for disseminating fake news is a possibility in this scenario. Applied research has been supported by promotional information as well as sensitive stuff over the network. Artificial Intelligence will be used to automatically assess the trustworthiness of social media accounts (AI). In this research, we describe a multilingual strategy to using Deep Learning to solve the bot identification problem on Twitter. End-users can utilise machine learning (ML) methodologies to assess the trustworthiness of a Twitter account. To achieve so, a number of tests were carried out using cutting-edge Multilingual Language Models. Construct an encoding of the user account's text-based features, which is then concatenated with the rest of the metadata to build a potential input vector on top of a Bot-DenseNet Dense Network. As a result, this article evaluates the language constraint from prior experiments where the encoding of the language was limited. Only the metadata information or the metadata information along with some other information was examined by the user account. properties of fundamental semantic text The Bot-DenseNet also generates a low-dimensional representation of the data. Within the Information Retrieval (IR) framework, a user account can be utilised for any application.

Protocol for Rhapsody: a longitudinal observational study examining the feasibility of speech phenotyping for remote assessment of neurodegenerative and psychiatric disorders

IntroductionNeurodegenerative and psychiatric disorders (NPDs) confer a huge health burden, which is set to increase as populations age. New, remotely delivered diagnostic assessments that can detect early stage NPDs by...

Extracting the Main Content of Web Pages Using the First Impression Area

Extracting the main content from a web page is essential in various applications such as web crawlers and browser reader modes. Existing extraction methods using text-based algorithms and features for English text can be ineffective for non-English web pages. This study proposes a main content extraction method that obtains visual and structural features from the rendered web page. Our method uses the first impression area (FIA), a part of a web page that users initially view. In this area, websites have applied many techniques that enable users to find the main content easily. Using the non-textual properties in the FIA, our method selects three points with high content area density and expands the area from each point until it meets several structural and visual-based conditions. We evaluated our method, browsers’ (Mozilla Firefox and Google Chrome) reader modes, and existing main content extraction methods on multilingual datasets using two measures: Longest Common Subsequences and matched text blocks. The results showed that our method performed better than other methods in both English (up to 46%, matched text blocks F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0.5</sub> ) and non-English (up to 42%, matched text blocks F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0.5</sub> ) web pages.

Caviar-Sunflower Optimization Algorithm-Based Deep Learning Classifier for Multi-Document Summarization

AbstractThis paper proposes a multi-document summarization model using an optimization algorithm named CAVIAR Sun Flower Optimization (CAV-SFO). In this method, two classifiers, namely: Generative Adversarial Network (GAN) classifier and Deep Recurrent Neural Network (Deep RNN), are utilized to generate a score for summarizing multi-documents. Initially, the simHash method is applied for removing the duplicate/real duplicate contents from sentences. Then, the result is given to the proposed CAV-SFO based GAN classifier to determine the score for individual sentences. The CAV-SFO is newly designed by incorporating CAVIAR with Sun Flower Optimization Algorithm (SFO). On the other hand, the pre-processing step is done for duplicate-removed sentences from input multi-document based on stop word removal and stemming. Afterward, text-based features are extracted from pre-processed documents, and then CAV-SFO based Deep RNN is introduced for generating a score; thereby, the internal model parameters are optimally tuned. Finally, the score generated by CAV-SFO based GAN and CAV-SFO based Deep RNN is hybridized, and the final score is obtained using a multi-document compression ratio. The proposed TaylorALO-based GAN showed improved results with maximal precision of 0.989, maximal recall of 0.986, maximal F-Measure of 0.823, maximal Rouge-Precision of 0.930, and maximal Rouge-recall of 0.870.