The prominence of social media poses a significant threat to information integrity as the spread of fake news increases. It becomes imperative for online media outlets to develop effective strategies to mitigate the spread of fake news. In this research, the PLNCC dataset is introduced as an expansion of two state-of-the-art fake news datasets. The objective is to improve the classification of fake news by extracting additional linguistic and psychological features, as well as user comment data. In this work, a quantitative analysis of the linguistic and psychological features of fake news articles and related user comments is performed. The efficacy of the PLNCC dataset is demonstrated through rigorous evaluation, showcasing its performance against state-of-the-art benchmark datasets. The classification models running on this dataset achieved a significant performance improvement, up to 10%, when compared to the original two datasets.

In this paper, we offer a new visualization tool -- Dataset Statistical View (DSV), to lower the barrier of research entry by providing easy access to the question-answering (QA) datasets that researchers can build their work upon. Our target users are new researchers to the QA domain with no prior knowledge nor programming skills. The system is populated with multiple QA datasets, which covers a wide range of QA tasks. It allows researchers to explore and compare existing QA datasets at a one-stop website. The system shows statistical graphs for each QA dataset to offer an overview and a visual comparison between datasets. Although this paper focuses mainly at the syntactic level comparison, integrating bias and semantic level analysis is our ongoing work. We believe our DSV system is a valuable contribution to the advancement of the QA field, as it provides a solid starting point for new researchers and practitioners. An overview of the framework is demonstrated in this paper and the introduction of the application system is available at https://cnchuy.github.io/images/demo.mp4.

These days, due to the advancement of information technology, recommendation system has become one of the key tools for e-commerce business. E-commerce platforms allow users to provide feedback in both written comments and numerical ratings. Recommendation systems are utilized to recommend users to new or unseen items based on these previously collected comments or ratings. In recent years, multi-aspect or multi-criteria based recommendation systems have been studied a lot by the recommendation research community. However, these research works are conducted either with reviews or ratings, not with both. In this project, we argue that integrating both textual reviews (with multiple aspects) and numerical multi-criteria ratings can further enhance the overall rating prediction accuracy. We propose a Multi-criteria Rating and Review based Recommendation model (MRRRec). We show that incorporating multi-criteria ratings into multi-aspect ratings from reviews has a great impact on performance. Our proposed model outperforms several state-of-the-art models such as ANR, DeepCoNN, and Deep Multi-criteria Recommendation System in terms of MSE, MAE, precision, recall, and F1. We show that our proposed model achieves an average of 19% and 23.0% lower MSE and MAE respectively and 7.0%, 1.0% and 3.8% higher precision, recall, and F1 score respectively. We further show that our model performs significantly better with Word2Vec word embedding than the GloVe word embedding method.

The news recommender systems are marked by a few unique challenges specific to the news domain. These challenges emerge from rapidly evolving readers’ interests over dynamically generated news items that continuously change over time. News reading is driven by a blend of a reader’s long-term and short-term interests. In addition, diversity is required in a news recommender system to keep the reader engaged in the reading process and get them exposed to different views and opinions. This paper proposes a deep neural network that jointly learns news and user representation in a unified framework. It learns the news representation (features) from the headlines, snippets (body) and taxonomy (category, subcategory) of news. The attention mechanism learns a reader’s long-term interests from the complete click history, short-term interests from recent clicks via LSTMs and diverse interests. We also apply different levels of attention to our model. We conduct extensive experiments on two news datasets to demonstrate the effectiveness of our approach.

The user feedback data such as likes, dislikes, comments on movie trailers posted on YouTube can be a useful information source for movie recommender systems. In this paper, we study the effect of adding the feedback data on trailers as a type of the side information to the movie rating data. We propose a recommendation framework that can integrate the trailer and rating data adopting different integration strategies: integrating all the trailer data as movie features, using sentiment scores derived from the trailer comments as a rating matrix to integrate with the movie rating matrix and treating others as the movie features, or only integrating the sentiment score based rating matrix with the movie rating matrix. Our experiment shows that if we include the movie trailer data, recommendation accuracy is improved. We also find that the most accurate result is achieved if all the trailer feedback data is integrated as movie features. To design our system, we use both Matrix Factorization (MF) and Deep Neural Network (DNN) Models. We find that the DNN model performs better than the MF model.

News recommender systems are usually designed to provide accurate and personalized recommendations to the readers. The diversity of the recommended results has received much less attention in this field. When it is considered, the current state-of-the-art models often apply the re-ranking mechanisms to promote the diversified results to the individual users. In this work, we propose a latent factor model to achieve the requisite level of accuracy while maintaining a reasonable level of diversity in a news recommender system. The existing latent factor methods mostly rely on Tikhonov regularization to improve the generality of the learnt models. These methods tend to focus mainly on accuracy measures, i.e., generating recommendations highly aligned with a user's past preference, which may cause a decrease in the diversity of information to which news readers are exposed. In our work, we make effective use of elastic-net regression to regularize the model for both the accuracy and the diversity in a single optimization framework. We demonstrate the effectiveness of our model over the state-of-the-art methods by conducting extensive experiments on a real-world news dataset.

In the past, news recommender systems have been built to recommend list of news items similar to those that a user has accessed before (content-based); or similar to those that have been read by similar users (collaborative filtering). However, the highly volatile nature of the news content and the dynamic and evolving user preferences are either ignored or not taken into full consideration in these systems. In a news recommender system, it is very likely that a user's short-term interest or preference may have a sudden change due to an emerging social or personal event or breaking news while their long-term interests may change gradually or remain. For these long-term interests of the readers, it is often more appropriate to associate them with news categories than with individual news items. In this paper, we propose a biased matrix factorization model with consideration of both temporal dynamics of user preferences and news taxonomy to build a news recommender system. By conducting an extensive experiment on a collection of news data, we demonstrate the effectiveness of our proposed model against traditional matrix factorization models as well as other neural recommender baselines. The findings from our experiments show that news category is an important factor when readers choose news articles to read, and temporal factors with consideration of different temporal resolution also play a role in this process.

Social network is a hot topic of interest for the researchers in the field of computer science in recent years. The vast amount of data generated by these social networks play a very important role in information diffusion. Social network data are generated by its users. So, user's behavior and activities are being investigated by the researchers to get a logical view of social network platform. This research proposed a novel retweet prediction model which considers difference in user's behavior as an author (as reflected in the tweets) and a retweeter (as reflected in the retweets) and do the prediction accordingly. The proposed retweet prediction strategy taking this difference into consideration, gave better prediction accuracy than the conventional strategy. The findings of this research explains that in social networks, some users show different behavior indifferent roles and these differences may have impact on future research.

Recently, researchers have taken tremendous strides in attempting to synthesize conventional social judgments and automated filtering within recommender systems. In this study, we aim to enhance recommendation efficiency via integrating social networking information with traditional recommendation algorithms. To achieve this objective, we first propose a new user similarity metric that not only considers tagging activities of users, but also incorporates their social relationships, such as friendship and membership, in measuring the closeness of two users. Subsequently, we define a new item prediction method which makes use of both user-to-user similarity and item-to-item similarity. Experimental outcomes on Last.fm show some positive results that attest the efficiency of our proposed approach.

In this paper, we investigate selection strategies for rescue servers in a fully collaborative in which heterogeneous systems share spare resource to address hotspot problems experienced by individual members. Different from content replication, delays due to application replication and startup time of the resulting rescue servers will be taken into consideration. Details about our replication strategy, such as the maximum set size margin on the total number of discarded requests, overall latency for enabling rescue servers, are given. Simulation result shows that our selection strategies perform much better than existing mechanisms in which replication and startup delay are ignored.

Normally, the unit for a ranking model in a Web IR system is a Web page, which is, sometimes, just an information fragment. A larger unit considering the linkage information may be desired to reduce the cognitive overload for users to identify the complete information from the interconnected Web. We propose a ranking model to measure the relevance of the whole Web site. We take some illustrations to show the idea and provide evidences to indicate its effectiveness.

No abstract available.

No abstract available.

It is observed that a better approach to Web information understanding is to base on its document framework, which is mainly consisted of (i) the title and the URL name of the page, (ii) the titles and the URL names of the Web pages that it points to, (iii) the alternative information source for the embedded Web objects, and (iv) its linkage to other Web pages of the same document. Investigation reveals that a high percentage of words inside the document framework are “compound words” which cannot be understood by ordinary dictionaries. They might be abbreviations or acronyms, or concatenations of several (partial) words. To recover the content hierarchy of Web documents, we propose a new word segmentation and recognition mechanism to understand the information derived from the Web document framework. A maximal bi-directional matching algorithm with heuristic rules is used to resolve ambiguous segmentation and meaning in compound words. An adaptive training process is further employed to build a dictionary of recognisable abbreviations and acronyms. Empirical results show that over 75% of the compound words found in the Web document framework can be understood by our mechanism. With the training process, the success rate of recognising compound words can be increased to about 90%.