Online reviews have a decisive impact on consumers’ purchasing decisions This opens the doors for spammers and scammers to post fake reviews for promoting non-existent products or undermine competitor products to affect social behavior Thus, the identification of reviews as fake and real has become ever more important Traditional approaches for text classification use a bag-of-words model to represent text which causes sparsity and word representations learnt from neural networks with limited ability to handle unknown words In this paper, we propose a technique based on three different models trained on the idea of a multi-view learning technique and create an ensemble of all models by employing an aggregation technique for generating final predictions The core idea of our methodology is to extract rich information from the text of reviews by combining bag-of-n-grams and parallel convolution neural networks(CNNs) By using an n-gram embedding layer with small kernel sizes we can use local context with the same computation power as required to train deep and complex CNNs Our CNN-based architecture consumes n-gram embeddings as input and uses the parallel convolutional blocks to extract richer feature representations from text Our approach for the detection of fake reviews also combines textual linguistic features and non-textual features related to reviewer behavior We evaluate our approach on publically available Yelp Filtered Dataset and achieve F1 scores of up to 92% for classifying fake reviews

Fake reviews classification using deep learning ensemble of shallow convolutions

Social media, today, demonstrates the rapid growth of modern society as it becomes the main platform for Internet users to communicate and express themselves. People around the world, use a number of devices and resources to access the Internet, set up social networks, conduct online business, e-commerce, e-surveys, etc. Currently, social media is not only a technology that provides information to consumers, it also encourages users to connect and share their views and perspectives. It leads to an increase in inspiration towards Opinion Mining (OM), which is important for both customers and companies in making decisions. Individuals like to see the opinions provided by other customers about a particular product or a service. Companies need to analyze their customer's feedback to strengthen their business decisions. A lot of research has been performed in various languages in the field of Aspect Based OM (ABOM). However, there are still certain languages that need to be explored, such as Roman Urdu (RU). This paper presents a proposed reviews data-set (a RU data-set) of mobile reviews that has been manually annotated with multi-aspect sentiment labels at the sentence-level. It presents base-line results using different Machine Learning (ML) algorithms. The results demonstrate 71% F1-score for aspect detection and 64% for aspect-based polarity.

Roman Urdu reviews dataset for aspect based opinion mining

Human relationships are influenced by the underlying emotions in their interactions. With the increasing use of social networks, relationships from textual data can also be inferred from online interactions. Such interactions result in massive amount of textual data which is available in the form of text messages, emails, and social media posts. Identification and analysis of human relationships are useful for numerous applications ranging from cybersecurity to public health. In this paper, we present a method called RIEA (Relationship Identification using Emotion Analysis), for identifying relationships between multiple intelligent agents by analyzing the conversation between them. The objective of our work is to combine concepts of cognitive psychology and natural language processing (NLP) to extract emotions and map them onto a set of relationships and analyze how relationships transform over time. We employ psychological models to label a large corpus of conversations and apply machine learning techniques to determine emotion-to-relationship mapping. We use four distinct association classes and four attachment styles using best-worst scaling method for classification. Combining the attachment and association styles given in research literature gives us the relationship combinations for our analysis. Additionally, this work studies the most common changes of behaviors and emotions and the corresponding transformations in human relationships. Our results show that RIEA can correctly detect interpersonal relationships with an accuracy of 85%. The evaluation shows that RIEA can accurately identify interpersonal relationships from conversations and can be extended for identifying more complex relationships. This study also highlights the effect of changes in emotional behavior in the development of relationships over time.

Relationship Identification Between Conversational Agents Using Emotion Analysis

Clickbait is an elusive challenge with the prevalence of social media such as Facebook and Twitter that misleads the readers while clicking on headlines. Limited annotated data makes it onerous to design an accurate clickbait identification system. The authors address this problem by purposing deep learning-based architecture with external knowledge which trains on social media post and descriptions. The pre-trained ELMO and BERT model obtains the sentence level contextual feature as knowledge; moreover, the LSTM layer helps to prevail the word level contextual feature. Training has done at different experiments (model with EMLO, model with BERT) with different regularization techniques such as dropout, early stopping, and finetuning. Forward context-aware clickbait tweet identification system (FCCTI) with BERT finetuning and model with ELMO using glove pre-trained embedding is the best model and achieves a clickbait identification accuracy of 0.847, improving on the previous baseline for this task.

Forward Context-Aware Clickbait Tweet Identification System

Pretrained language models are now of widespread use in Natural Language Processing. Despite their success, applying them to Low Resource languages is still a huge challenge. Although Multilingual models hold great promise, applying them to specific low-resource languages e.g. Roman Urdu can be excessive. In this paper, we show how the code-switching property of languages may be used to perform cross-lingual transfer learning from a corresponding high resource language. We also show how this transfer learning technique termed Bilingual Language Modeling can be used to produce better performing models for Roman Urdu. To enable training and experimentation, we also present a collection of novel corpora for Roman Urdu extracted from various sources and social networking sites, e.g. Twitter. We train Monolingual, Multilingual, and Bilingual models of Roman Urdu - the proposed bilingual model achieves 23% accuracy compared to the 2% and 11% of the monolingual and multilingual models respectively in the Masked Language Modeling (MLM) task.

Bilingual Language Modeling, A transfer learning technique for Roman Urdu.

Social networks are generating huge amounts of complex textual data which is becoming increasingly difficult to process intelligently. Misinformation on social media networks, in the form of fake news, has the power to influence people, sway opinions and even have a decisive impact on elections. To shield ourselves against manipulative misinformation, we need to develop a reliable mechanism to detect fake news. Yellow journalism along with sensationalism has done a lot of damage by misrepresenting facts and manipulating readers into believing false narratives through hyperbole. Clickbait does exactly this by using characteristics of natural language to entice users into clicking a link and can hence be classified as fake news. In this paper, we present a deep learning framework for clickbait detection. The framework is trained to model the intrinsic characteristics of clickbait for knowledge discovery and then used for decision making by classifying headlines as either clickbait or legitimate news. We focus our attention on the linguistic analysis during the knowledge discovery phase as we investigate the underlying structure of clickbait headlines using our Part of Speech Analysis Module. The decision-making task of classification is carried out using long short-term memory. We believe that it is our framework’s architecture that has played a pivotal role to outperform the current state of the art with a classification accuracy of 97%.

Aymen Khan

Papers

A deep learning framework for clickbait detection on social area network using natural language cues