March 11 ~ 12, 2023, Virtual Conference
Danqin Wu, Department of Computer Science, Beijing University of Posts & Telecommunications, Beijing, China
In multi-turn dialogue generation, responses are not only related to the topic and background of the context but also related to some words and phrases in the sentences of the context. However, currently widely used hierarchical dialog models solely rely on context representations from the utterance-level encoder, ignoring the sentence representations output by the word-level encoder. This inevitably results in a loss of information while decoding and generating. In this paper, we propose a new dialog model XReCoSa with method X-Fusion to tackle this problem which aggregates multi-scale context information for hierarchical dialog models. Firstly, the hierarchical dialogue model uses a word-level encoder and an utterance-level encoder respectively to encode context to obtain sentence representations and context representations. Then, X-Fusion adopts a Multi-head attention module in the generation decoder to fuse high-level semantics in contexts into the low layers of the decoder, and subsequently pay attention to lowlevel sentence representations at the high layers of the decoder, and finally generate the response. we conduct experiments on the English dataset DailyDialog. Experimental results exhibit that our method outperforms baseline models on both automatic metric-based and human-based evaluations.
Dialogue Generation, Self-Attention, Multi-Turn Dialogue & Context Awareness.
Muhammad Danial Khilji
The feature matching is a basic step in matching different datasets. This article proposes shows a new hybrid model of an advanced NLP model (BERT model) used in parallel with a statistical model based on Jaccard similarity to measure the similarity between list of features from two different datasets. This reduces the time required to search for correlations or manually match each feature from one dataset to another.
BERT, Cosine similarity, features, matching, semantic similarity, similarity.
Pei Wang, Zhen Guo, Lifu Huang, and Jin-Hee Cho, Department of Computer Sciences, Virginia Tech, Falls Church, VA, USA
This paper deals with an important topic of cybergrooming and sexual misconducts in artificial intelligence (AI)-based educational programs. Although cybergrooming (or online sexual exploitation) has been recognized as a serious cybercrime, there have been insucient programs to proactively protect the youth from ybergrooming. In this work, we present a generative chatbot framework, called SERI (Stop cybERgroomIng), that can generate fluent authentic conversations in the context of cybergrooming between a perpetrator chatbot and a potential victim chatbot. Furthermore, we propose deep reinforcement learning (DRL)-based dialogue eneration with a stage-related reward to lead the conversation to the expected stage. SERI is designed to strengthen the youth's precautionary awareness of cybergrooming in a safe and authentic environment. At the same time, we aim to minimize any potential ethical issues that may be introduced by using perverted languages when the eveloped chatbots are deployed to youth for cybersecurity education programs. We evaluated the quality of conversations generated by SERI based on open-source referenced, unreferenced metrics, and human evaluation. We developed SERI that would be used as a platform of deploying the perpetrator chatbot to interact with human users (i.e., youth) to observe youth users' responses to strangers or acquaintances and collect the reactions when the youth users are asked for private or sensitive information by the perpetrator.
Cybergrooming, natural language processing, deep reinforcement learning, chatbots.
Anza Shadrach Peter, West African Examinations Council, Nigeria
The easiness of communication remains the most efficient means or tool towards making the World a global Village. With over 5000 spoken languages across the Universe, the oldest and most natural means of Communication remains Speech or language. Hence, the idea of a Speech recognition system in today’s World need not be overemphasized. Intuitively, this system performs the recognition of speech samples in the three major Nigerian languages (Hausa, Igbo and Yoruba). Mel Frequency Cepstral Coefficients (MFCC) is used to extract features from the speech signals of spoken words. Furthermore, Neural Network was used to train and test the audio files to get the recognized samples. Classification and recognition of the Speech signals was achieved with an accuracy rate of 70%.
Translation,MFCC, Neural Network, Speech Recognition
Yu Liu1, Xue Qiu2 and Kai Wang3, 1Department of Software Engineering, Dalian University of Technology,Dalian, China 2Department of Software Engineering, Dalian University of Technology,Dalian, China 3School of Computer Science and Engineering, Nanyang Technological University,Singapore
This study aims to verify the reliability and availability of Semantic Scholar (SS), and compare with three popular databases, including Scopus, Microsoft Academic (MA), and Web of Science (WoS). This paper mainly analyzes from the functional level and data level. At the functional level, the new functions provided by Semantic Scholar are analyzed in the aspects of the search mode, academic influence indicators, open-source, and other aspects. We also carry out statistical analysis of the full database of Semantic Scholar. At the data level, we construct a dataset that is composed of computer science articles in 50 journals from 2016 to 2019. And we conduct statistical analysis from the perspectives of journals, article coverage rate, paper type. In addition, we focus on comparing the citation number and reference distribution of different data sources.
Bibliometric Analysis, Functional comparison, Full database analysis, Citationanalysis
Zainab Mansur1, Nazlia Omar2 and Sabrina Tiun3, 1Department of Computer Science, Faculty of Sciences, Omar Al-Mukhtar University, Al Bayda, Libya 1Center for Artificial Intelligence Technology, Universiti Kebangsaan Malaysia, Bangi Malaysia 1Center for Artificial Intelligence Technology, Universiti Kebangsaan Malaysia, Bangi Malaysia
Informal methods of communication, like tweets, rely heavily on initialization abbreviations to reduce message size and time, making them difficult to mine and normalize using existing methods. Therefore, this present study compiled a lexicon repository to normalize the initialism abbreviations used in tweets in the English language. Several components were taken into consideration while compiling the repository. This included the Tweepy Python library, keyword list, small developed rules, and online dictionaries. A lexicon repository of 300 abbreviations and their complete forms was compiled. This will be used in an ongoing study to normalize Twitter hate speech data and to detect it.
normalization, hate speech, Twitter, abbreviation, dictionary
Jawid Ahmad Baktash, Mursal Dawodi LIA, Avignon University Avignon, France
Nowadays, text classification for different purposes becomes a basic task for concerned people. Hence, much research has been done to develop automatic text classification for the majority of national and international languages. However, the need for an automated text classification system for local languages is felt. The main purpose of this study is to establish a novel automatic classification system of Pashto text. In order to follow this up, we established a collection of Pashto documents and constructed the dataset. In addition, this study includes several models containing statistical techniques and neural network neural machine learning including DistilBERT-base-multilingual-cased, Multilayer Perceptron, Support Vector Machine, K Nearest Neighbor, decision tree, Gaussian naïve Bayes, multinomial naïve Bayes, random forest, and logistic regression to discover the most effective approach. Moreover, this investigation evaluates two different feature extraction methods including bag of words, and Term Frequency Inverse Document Frequency. Subsequently, this research obtained an average testing accuracy rate of 94% using the MLP classification algorithm and TFIDF feature extraction method in single label multi-class classification. Similarly, MLP+TFIDF with F1-measure of 0.81 showed the best result. Experiments on the use of pre-trained language representation models (such as DistilBERT) for classifying Pashto texts show that we need a specific tokenizer for a particular language to obtain reasonable results.
Pashto, DistilBERT, BERT, Multi-lingual BERT, Multi-layer Perceptron, Support Vector Machine, K Nearest Neighbor, Decision Tree, Random Forest, Logistic Regression, Gaussian Naïve Bayes, Multinomial Naïve Bayes, TFIDF, Unigram, Deep Neural Network, Classification
Mursal Dawodi, Jawid Ahmad Baktash LIA, Avignon University Avignon, France
Recently, many researchers have focused on building and improving speech recognition systems to facilitate and enhance human-computer interaction. Today, Automatic Speech Recognition (ASR) system has become an important and common tool from games to translation systems, robots, and so on. However, there is still a need for research on speech recognition systems for low-resource languages. This article deals with the recognition of a separate word for Dari language, using Mel-frequency cepstral coefficients (MFCCs) feature extraction method and three different deep neural networks including Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Multilayer Perceptron (MLP), and two hybrid models of CNN and RNN. We evaluate our models on our built-in isolated Dari words corpus that consists of 1000 utterances for 20 short Dari terms. This study obtained the impressive result of 98.365% average accuracy.
Dari, deep neural network, speech recognition, recurrent neural network, multilayer perceptron, convolutional neural network
Jiachun Chen1 and Zhaogong Zhang2, 1Department of Computer Science and Technology, Heilongjiang University, HeiLongJiang, China 2Heilongjiang University, HeiLongJiang, China
As a key task in entity linking, entity disambiguation is of great significance to entity linking. Entity data on the network usually has phenomena such as polysemy or aliases. Computers cannot directly identify and use these entity data.Due to the existence of multiple interpretations, these entities cannot uniquely determine the target link entity. Usually, entity semantics, properties, the association between features or enti ties to disambiguate the referential ambiguity between entities.However, most of the current Graph Neural Network-based algorithms are limited by the Knowledge Graph of the Knowledge Base constrution, and there are few association paths between entities, resulting in very sparse graphs, so that the hidden relationships between entities have not been learned and cannot be obtained better effect.In the process of entity disambiguation, this paper vectorizes the text description and knowledge base entity abstract based on the Bert pre-training model, so as to classify the entity by Clsf, and embed the entity property information with the help of CBOW to fully preserve the entity property information, which reduces the computational cost. while improving the accuracy. The experimental results show that, compared with the traditional model, the model performs better on the CCKS2019 dataset.
Entity disambiguation，Bert model， Knowledge Base，entity property
Samin Poudel1and Marwan Bikdash1, 1Department of Computational Data Science and Engineering, North Carolina A & T University, Greensboro, USA
The performance of a Collaborative Filtering (CF) method is based on the properties of a User-Item Rating Matrix (URM). And the properties or Rating Data Characteristics (RDC) of a URM are constantly changing. Recent studies significantly explained the variation in the performancesof CF methods resulted due to the change in URM using six or more RDC. Here, we found that the significant proportion of variation in the performances of different CF techniques can be accounted to two RDC only. The two RDC are the number of ratings per user or Information per User (IpU) and the number of ratings per item or Information per Item (IpI). Andthe performances of CF algorithms are quadratic to IpU (or IpI) for a square URM. The findings of this study are based on seven well-established CF methods and three popular public recommender datasets: 1M MovieLens, 25M MovieLens, and Yahoo! Music Rating datasets.
Rating Matrix, Recommendation System,Collaborative Filtering, Data Characteristics
Asma BelHadjBraiek, ZouhourNeji Ben Salem, Carthage university, Tunisia Faculty of Economic Sciences and Management of Nabeul university campus, ElMrezgua, 8000
Social networks are the most used means to express oneself freely and give one's opinion about a subject, an event or an object. These networks present a rich content that could be subject today to sentiment analysis interest in many fields such as politics, social sciences, marketing and economics. However, social networks users’ express themselves using their own dialect. Thus, in order to help decision makers in the analysis of users' opinions, it is necessary to proceed to the sentimental analysis of this dialect. Paper subject deals with a hybrid model combining lexicon-based approach with a modified and adapted version of sentiment rule-based engine named VADER. The hybrid model is tested and evaluated using Tunisian Arabic Dialect, it showed good performance reaching 85% classification.
automatic language processing, sentiment analysis, text mining, emotional detection, social web, annotated corpus, sentiment lexicon, sentiment engine.