TRUST IN DECONSTRUCTED RECOMMENDER SYSTEMS. CASE STUDY: NEWS RECOMMENDER SYSTEMS

A significant increase in news sources has been brought about by the growth of the online and the social web. The easy access to numerous news sources results in a flood of information that is frequently contradictory and confusing. When news circulates on the social web, it can be challenging to distinguish between legitimate and false reports. False news can be used for a variety of purposes, including as political influence, financial gain, and strict convictions to solve these problems, a new news aggregator was developed to help users distinguish between authentic, fake, and copy content. News from multiple news sources is reviewed, approved, and presented to the client in the suggested system. The news aggregator also suggests important news to the customer by anticipating consumer preference via a recommendation algorithm. It has been demonstrated that news recommendation systems can naturally handle lengthy articles and distinguish between them for readers who are taking into account established models.

This study aims to explore the impact of personalized news recommendation systems based on big data and artificial intelligence algorithms on audience cognition and behavior. With the rapid development of information technology, news recommendation systems have been widely used in the media industry, but their impact on audiences has not been fully studied. This paper adopts a mixed research method, combining questionnaire surveys, in-depth interviews and experimental research, and conducts a three-month follow-up survey on 500 news app users. The results show that the personalized recommendation system significantly increases users' news reading volume and retention time, but also leads to the intensification of information cocoon effect and cognitive bias. In addition, the study finds that the transparency of the recommendation algorithm, the user's media literacy level and the diversity of news sources are key factors affecting audience cognition and behavior. Based on the research findings, this paper proposes suggestions for optimizing the news recommendation system, improving users' media literacy and promoting information diversification, in order to minimize the negative impact of personalized recommendations while improving user experience. This study has important theoretical and practical significance for understanding the characteristics of audience behavior in the intelligent media environment, improving the design of news recommendation systems and formulating relevant policies.

Recent rises in political polarization across the globe are often ascribed to algorithmic content filtering on social media, news platforms, or search engines. The widespread usage of news recommendation systems (NRS) is theorized to drive users in homogenous information environments and, thereby, drive affective, ideological, and perceived polarization. To test this assumption, we conducted an online experiment ( n = 750) with running algorithms that enriches content-based NRS with negative or neutral sentiment. Our experiment finds only limited evidence for polarization effects of content-based NRS. Nevertheless, the time spent with an NRS and its recommended articles seems to play a crucial role as a moderator of polarization. The longer participants were using an NRS enriched with negative sentiment, the more they got affectively polarized, whereas participants using an NRS incorporating balanced sentiment ideologically depolarized over time. Implications for future research are discussed.

News recommender systems rely on rich textual information from news articles to generate user-specific recommendations. This reliance may expose these systems to potential vulnerabilities through textual attacks. To explore this vulnerability, we propose LANCE, a LArge language model-based News Content rEwriting framework, designed to influence news rankings and highlight the unintended promotion of manipulated news. LANCE consists of two key components: an explorer and a reflector. The explorer first generates rewritten news using diverse prompts, incorporating different writing styles, sentiments, and personas. We then collect these rewrites, evaluate their ranking impact within news recommender systems, and apply a filtering mechanism to retain effective rewrites. Next, the reflector fine-tunes an open-source LLM using the successful rewrites, enhancing its ability to generate more effective textual attacks. Experimental results demonstrate the effectiveness of LANCE in manipulating rankings within news recommender systems. Unlike attacks in other recomendation domains, negative and neutral rewrites consistently outperform positive ones, revealing a unique vulnerability specific to news recommendation. Once trained, LANCE successfully attacks unseen news recommender systems (i.e., those for which LANCE received no information during training), highlighting its generalization ability and exposing shared vulnerabilities across different systems. Our work underscores the urgent need for research on textual attacks and paves the way for future studies on defense strategies.

Recommender systems are widely applied in digital platforms such as news websites to personalize services based on user preferences. In news websites most of users are anonymous and the only available data is sequences of items in anonymous sessions. Due to this, typical collaborative filtering methods, which are highly applied in many applications, are not effective in news recommendations. In this context, session-based recommenders are able to recommend next items given the sequence of previous items in the active session. Neighborhood-based session-based recommenders has been shown to be highly effective compared to more sophisticated approaches. In this study we propose scenarios to make these session-based recommender systems diversity-aware and to address the filter bubble phenomenon. The filter bubble phenomenon is a common concern in news recommendation systems and it occurs when the system narrows the information and deprives users of diverse information. The results of applying the proposed scenarios show that these diversification scenarios improve the diversity measures in these session-based recommender systems based on four news datasets.

Recommender systems are becoming an essential part of smart services. When building a news recommender system, we should consider special features different from other recommender systems. Hot news topics are changing every moment, thus it is important to recommend right news at the right time. This paper aims to propose a new model, based on deep neural network, to analyse user preference for news recommender system. The model extracts interest keywords to characterize the user preference from the set of news articles read by that particular user in the past. The model utilizes characterizing features for news recommendation, and applies those to the keyword classification for user preference. For the keyword classification, we use deep neural network for online preference analysis, because adaptive learning is necessary to track changes of hot topics sensitively. The usefulness of our model is validated through experiments. In addition, the accuracy and diversity of the recommendation results is also analysed.

Recommender systems are designed to help users quickly access large volumes of information according to their profiles. Most previous works in recommender systems have put their emphasis on the accuracy of finding the most similar items according to a user’s profile, while often ignoring other aspects that may affect users’ experiences in practice, e.g., the novelty and diversity issues within a recommendation list. In this paper, we focus on utilizing taxonomic knowledge extracted from an online encyclopedia to boost a content-based personalized news recommender system without much human involvement. Given a recommendation list, we improve a user’s satisfaction by introducing the taxonomy based novelty and diversity metrics to include novel, but potentially related items into the list, and filter out redundant ones. The experimental results show that the coarse grained knowledge resources can help a content-based news recommender system provides accurate as well as user-oriented recommendations.

How do users interact with algorithm recommender systems? The interaction of users, algorithms, and performance

Numerous recommendation approaches are in use today. However, comparing their effectiveness is a challenging task because evaluation results are rarely reproducible. In this article, we examine the challenge of reproducibility in recommender-system research. We conduct experiments using Plista's news recommender system, and Docear's research-paper recommender system. The experiments show that there are large discrepancies in the effectiveness of identical recommendation approaches in only slightly different scenarios, as well as large discrepancies for slightly different approaches in identical scenarios. For example, in one news-recommendation scenario, the performance of a content-based filtering approach was twice as high as the second-best approach, while in another scenario the same content-based filtering approach was the worst performing approach. We found several determinants that may contribute to the large discrepancies observed in recommendation effectiveness. Determinants we examined include user characteristics (gender and age), datasets, weighting schemes, the time at which recommendations were shown, and user-model size. Some of the determinants have interdependencies. For instance, the optimal size of an algorithms' user model depended on users' age. Since minor variations in approaches and scenarios can lead to significant changes in a recommendation approach's performance, ensuring reproducibility of experimental results is difficult. We discuss these findings and conclude that to ensure reproducibility, the recommender-system community needs to (1) survey other research fields and learn from them, (2) find a common understanding of reproducibility, (3) identify and understand the determinants that affect reproducibility, (4) conduct more comprehensive experiments, (5) modernize publication practices, (6) foster the development and use of recommendation frameworks, and (7) establish best-practice guidelines for recommender-systems research.

An increasing number of media organisations are adopting news recommender systems (NRS). Such algorithmic technologies, which prioritise content based on, for example, previous user behaviour or popularity metrics, may have far-reaching repercussions for news work. Despite this, the implications of NRS implementation for intra-organisational practices as well as dynamics and tensions between involved actors remain understudied. Against this background, this study examines decision-making processes and relationships between actors participating in NRS projects from an institutional logics perspective and places a particular emphasis on resulting tensions between journalistic, market, and tech logics. Drawing on 32 in-depth qualitative interviews with news media professionals across ten news organisations in the Netherlands and Switzerland, we discover a wide range of strategies which aim to reconcile logic multiplicity in the specific case of NRS development. Such negotiation efforts can ultimately promote new work practices and forms of collaboration but may also have broader implications for the distribution of power and voice within news organisations.

With the continuous development of information technology, we are now in an era of information explosion, and we are also facing the problem of information overload, i.e., it is difficult to find the news that interests us from the huge news sources on the Internet. Based on this problem, this paper proposes a perceptron model-based news recommendation system, which uses the perceptron model to extract news keywords and realize the classification of the news; for user interest preferences, we analyze users’ news browsing preferences by collecting user behavior logs and use a multi-layer perceptron model to aggregate users’ interest features. The perceptron model is compared with traditional recommendation methods in terms of accuracy and recall, and one user is selected from 50 users who have been using the system for three months to analyze their news browsing preferences by single-day, short-term and long-term news data to verify the effectiveness of the perceptron model in aggregating users’ interest features. In terms of accuracy and recall, the accuracy of the perceptron model is higher than that of the traditional method for each news type, improving the accuracy by at least 5%-10%, e.g., 10% for real estate news and 9% for financial news. The recall rate of the perceptron model is improved by more than 6% for all news types except for social and entertainment news types. In terms of user interest preference mining, the whole change process of user browsing preference trend is in line with the human forgetting curve, which verifies that the perceptron model in the system is better used and meets the design requirements. Therefore, the applicability and reliability of the model proposed in this paper for news recommendation systems are verified in terms of accuracy, recall, and user interest preference mining.

A fusion sorting model based on logical regression is used to solve the problem of information overload in the news recommendation system and meet the personalized needs of users. First, the fusion sorting model is trained based on the article data and user data, and the original candidate news set is obtained by multi-channel recall method. Then the original candidate set is filtered by the fusion sorting model, and the recommendation list is generated. Finally, the experiment is carried out based on the Wuli News Recommendation System. Under the evaluation indexes of precision, recall and ILS, the fusion sorting model adopted significantly improves the recommendation ability of the system, and achieves the goal of making multiple single recommendation strategies make up for each other’s shortcomings.

More and more netizens prefer to comment on social hot issues today and their views become very useful for government decision-making. Specially, news and related comments often influence decision behavior of officers. However, it becomes a key problem to analyze them automatically in order to provide references for decision-making. One of effective way is to cluster news comments. In this paper, we discuss the k-means clustering algorithm and how to cluster news comments in order to obtain types of a special news comments. And we do an experiment on a real dataset collected from the news recommender system we developed for government decision-making. Primary results are shown that our k-means clustering method is effective and can be taken as an analysis method used in our recommender system.

Using complex language models has been a common strategy as personalized news recommendation systems are adopted by online news sites more and more. Before the epoch of GPT-3, news recommendation systems underwent a progression from rule-based and collaborative filtering approaches in the pre-2010s, through the integration of neural networks in the 2010s, to the emergence of earlier iterations of large language models like GPT-2 in 2019. Pre-trained language models have ushered in a new era of recommendation paradigms, thanks to the emergence of huge language models like GPT-3 and T-5. With its easy-to-use interface, ChatGPT is becoming more and more popular for text-based jobs. Focusing on news provider fairness, individualized news recommendations, and fake news identification, this study starts an inquiry into ChatGPT's efficacy in news recommendations. We acknowledge that ChatGPT's output sensitivity to input phrasing is a limitation, and our goal is to investigate these limitations from each angle. We also investigate whether certain prompt formats can help to alleviate these constraints or if more research is needed. To go beyond set assessments, we create a webpage where ChatGPT's performance on the examined activities and prompts is tracked once a week. Through the use of big language models, this work seeks to improve news recommendation performance and stimulate more research in this area.

In recent years, the type and quantity of news are growing rapidly, and it is not easy for users to find the news they are interested in the massive amount of news. A news recommendation system can score and predict the candidate news, and finally recommend the news with high scores to users. However, existing user models usually only consider users’ long-term interests and ignore users’ recent interests, which affects users’ usage experience. Therefore, this paper introduces gated recurrent unit (GRU) sequence network to capture users’ short-term interests and combines users’ short-term interests and long-term interests to characterize users. While existing models often only use the user’s browsing history and ignore the variability of different users’ interest in the same news, we introduce additional user’s ID information and apply the personalized attention mechanism for user representation. Thus, we achieve a more accurate user representation. We also consider the risk of compromising user privacy if the user model training is placed on the server side. To solve this problem, we design the training of the user model locally on the client side by introducing a federated learning framework to keep the user’s browsing history on the client side. We further employ secure multi-party computation to request news representations from the server side, which protects privacy to some extent. Extensive experiments on a real-world news dataset show that our proposed news recommendation model has a better improvement in several performance evaluation metrics. Compared with the current state-of-the-art federated news recommendation models, our model has increased by 0.54% in AUC, 1.97% in MRR, 2.59% in nDCG@5%, and 1.89% in nDCG@10. At the same time, because we use a federated learning framework, compared with other centralized news recommendation methods, we achieve privacy protection for users.