Peng Liu, Lemei Zhang, Terje Farup, Even W. Lauvrak, Jon Espen Ingvaldsen, Simen Eide, Jon Atle Gulla, Zhirong Yang. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

With the rapid advancement of Natural Language Processing in recent years, numerous studies have shown that generic summaries generated by Large Language Models (LLMs) can sometimes surpass those annotated by experts, such as journalists, according to human evaluations. However, there is limited research on whether these generic summaries meet the individual needs of ordinary people. The biggest obstacle is the lack of human-annotated datasets from the general public. Existing work on personalized summarization often relies on pseudo datasets created from generic summarization datasets or controllable tasks that focus on specific named entities or other aspects, such as the length and specificity of generated summaries, collected from hypothetical tasks without the annotators' initiative. To bridge this gap, we propose a high-quality, personalized, manually annotated summarization dataset called PersonalSum. This dataset is the first to investigate whether the focus of public readers differs from the generic summaries generated by LLMs. It includes user profiles, personalized summaries accompanied by source sentences from given articles, and machine-generated generic summaries along with their sources. We investigate several personal signals — entities/topics, plot, and structure of articles—that may affect the generation of personalized summaries using LLMs in a few-shot in-context learning scenario. Our preliminary results and analysis indicate that entities/topics are merely one of the key factors that impact the diverse preferences of users, and personalized summarization remains a significant challenge for existing LLMs.

Abstract The emergence of Pre-trained Language Models (PLMs) has achieved tremendous success in the field of Natural Language Processing (NLP) by learning universal representations on large corpora in a self-supervised manner. The pre-trained models and the learned representations can be beneficial to a series of downstream NLP tasks. This training paradigm has recently been adapted to the recommendation domain and is considered a promising approach by both academia and industry. In this paper, we systematically investigate how to extract and transfer knowledge from pre-trained models learned by different PLM-related training paradigms to improve recommendation performance from various perspectives, such as generality, sparsity, efficiency and effectiveness. Specifically, we propose a comprehensive taxonomy to divide existing PLM-based recommender systems w.r.t. their training strategies and objectives. Then, we analyze and summarize the connection between PLM-based training paradigms and different input data types for recommender systems. Finally, we elaborate on open issues and future research directions in this vibrant field.

Abstract Recent advances in graph-based learning approaches have demonstrated their effectiveness in modelling users’ preferences and items’ characteristics for Recommender Systems (RSs). Most of the data in RSs can be organized into graphs where various objects (e.g. users, items, and attributes) are explicitly or implicitly connected and influence each other via various relations. Such a graph-based organization brings benefits to exploiting potential properties in graph learning (e.g. random walk and network embedding) techniques to enrich the representations of the user and item nodes, which is an essential factor for successful recommendations. In this paper, we provide a comprehensive survey of Graph Learning-based Recommender Systems (GLRSs). Specifically, we start from a data-driven perspective to systematically categorize various graphs in GLRSs and analyse their characteristics. Then, we discuss the state-of-the-art frameworks with a focus on the graph learning module and how they address practical recommendation challenges such as scalability, fairness, diversity, explainability, and so on. Finally, we share some potential research directions in this rapidly growing area.

This paper presents a simple but effective method to build sentiment lexicons for the three Mainland Scandinavian languages: Danish, Norwegian and Swedish. This method benefits from the English Sentiwordnet and a thesaurus in one of the target languages. Sentiment information from the English resource is mapped to the target languages by using machine translation and similarity measures based on sentence embeddings. A number of experiments with Scandinavian languages are performed in order to determine the best working sentence embedding algorithm for this task. A careful extrinsic evaluation on several datasets yields state-of-the-art results using a simple rule-based sentiment analysis algorithm. The resources are made freely available under an MIT License.

A rapidly changing news ecosystem presents new challenges to research, media organizations, consumers, and societies. The 10th edition of the International Workshop on News Recommendation and Analytics (INRA) serves to exchange ideas and discuss recent trends, technological advancements, and open problems concerning news. We welcome contributions in scientific articles, demonstrations, and ideas. We strive to bring together researchers, practitioners, and decision-makers to address crucial challenges. The workshop provides an opportunity to learn about recent research and interactively discuss technical and interdisciplinary aspects related to news. Topics of interest include information access systems for news, advances in natural language processing, multi-modality, mis- and disinformation, trust and user experiences, and personalization.

News portals, social media, and news recommender systems have a strong influence on the perception of events. The way with which people engage with news has changed. Today, we encounter personalized access to news. On the one hand, personalization allows us to manage the overwhelming amount of information. On the other hand, personalization can create a set of problems such as filter bubbles, privacy and disinformation related problems. News analytics helps us to develop solutions towards the challenges created by personalization. News analytics helps us to understand the news ecosystem better and develop solutions towards the challenges of news recommender systems. The 9th International Workshop on News Recommendation and Analytics (INRA 2021) provides a forum to discuss recent trends and observations related to news recommendation, personalization, and analytics. The interdisciplinary workshop connects research from machine learning and analytics, algorithmic modelling and prediction, as well as results from ethical and psychological research.

Entity matching is the problem of identifying which records refer to the same real-world entity. It has been actively researched for decades, and a variety of different approaches have been developed. Even today, it remains a challenging problem, and there is still generous room for improvement. In recent years, we have seen new methods based upon deep learning techniques for natural language processing emerge. In this survey, we present how neural networks have been used for entity matching. Specifically, we identify which steps of the entity matching process existing work have targeted using neural networks, and provide an overview of the different techniques used at each step. We also discuss contributions from deep learning in entity matching compared to traditional methods, and propose a taxonomy of deep neural networks for entity matching.

With the dramatic expansion of international markets, consumers write reviews in different languages, which poses a new challenge for Recommender Systems (RSs) dealing with this increasing amount of multilingual information. Recent studies that leverage deep-learning techniques for review-aware RSs have demonstrated their effectiveness in modelling fine-grained user-item interactions through the aspects of reviews. However, most of these models can neither take full advantage of the contextual information from multilingual reviews nor discriminate the inherent ambiguity of words originated from the user’s different tendency in writing. To this end, we propose a novel Multilingual Review-aware Deep Recommendation Model (MrRec) for rating prediction tasks. MrRec mainly consists of two parts: (1) Multilingual aspect-based sentiment analysis module (MABSA), which aims to jointly extract aligned aspects and their associated sentiments in different languages simultaneously with only requiring overall review ratings. (2) Multilingual recommendation module that learns aspect importances of both the user and item with considering different contributions of multiple languages and estimates aspect utility via a dual interactive attention mechanism integrated with aspect-specific sentiments from MABSA. Finally, overall ratings can be inferred by a prediction layer adopting the aspect utility value and aspect importance as inputs. Extensive experimental results on nine real-world datasets demonstrate the superior performance and interpretability of our model.

Publishing news represents a vital function for societal health. News recommender systems, which support readers finding relevant content, face challenges beyond those encountered by other types of recommender systems. They have to deal with a dynamic flow of unstructured, fragmentary, and potentially unreliable news stories. The International Workshop on News Recommendation and Analytics (INRA) focuses on the challenges of news recommender systems and aims to connect researchers, practitioners and journalists. The seventh edition of INRA takes place as a half-day workshop in conjunction with thirteenth ACM Conference on Recommender Systems (RecSys '19) on September 16--20, 2019 in Copenhagen, Denmark. INRA 2019 focuses on the news recommender systems under three main categories: News recommendation, news analytics, and ethical aspects of news recommendation.

Decision-making in an emergency department needs to be efficient. It does not allow observation of the patient for a prolonged period of time, especially if the patients harm themselves or others, or refuses treatment. This includes suicidal, violent, intentional self-inflicted or non-consenting to treatments patient. Clinicians have to quickly decide whether to call the police, admit the patient to the psychiatric ward, according to recommended, predefined procedures. Since, for different situations different laws are applicable, clinicians need to follow certain procedures and consider specific legal issues while also handling challenging patients at the emergency department. Finding the right information quickly and following the recommended procedure is essential since the repercussions for a doctor acting incorrectly could be severe. This paper presents a mobile decision support system (DSS) prototype designed for healthcare providers to identify and follow the correct procedure based on applicable laws to mitigate the chance of incorrect actions being taken, along with the associated risks and preserve both the patients and their own rights. We explored healthcare providers requirements and alternatives for visualizing information by conducting observation, semi structured interviews and a total of four co-design workshops. The resulting mobile DSS enables clinicians to quickly access the correct information regarding their power and authority while also protecting the safety and welfare of the patient. Usability evaluation of the prototype showed that the average time spent to complete a given task was about 12 seconds by a total of six medical interns. Participants of the evaluation described the system easy to use, intuitive and highly useful for everyday use.

Time plays a crucial role in influencing and understanding users' changeable preferences among various factors which directly or indirectly result in users' interesting behavior. In this paper, we propose a neural time series forecasting model (NTSF) to fit and predict user preference trend according to time. In this model, we integrate emerging/hot topic detection to deal with the short-term aspects, and use Fast Fourier Transformation (FFT) to differentiate cyclic behavior of users. Considering the nonstationary and nonlinear characteristics appearing through user interest patterns, Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) is employed to balance the influence of short and long term aspects, as well as adjust model parameters according to historical results. Our experimental results with extensive Twitter datasets verify the effectiveness of our approach.

Datasets for recommender systems are few and often inadequate for the contextualized nature of news recommendation. News recommender systems are both time- and location-dependent, make use of implicit signals, and often include both collaborative and content-based components. In this paper we introduce the Adressa compact news dataset, which supports all these aspects of news recommendation. The dataset comes in two versions, the large 20M dataset of 10 weeks' traffic on Adresseavisen's news portal, and the small 2M dataset of only one week's traffic. We explain the structure of the dataset and discuss how it can be used in advanced news recommender systems.

This paper presents classification results for the analysis of sentiment in political news articles. The domain of political news is particularly challenging, as journalists are presumably objective, whilst at the same time opinions can be subtly expressed. To deal with this challenge, in this work we conduct a two-step classification model, distinguishing first subjective and second positive and negative sentiment texts. More specifically, we propose a shallow machine learning approach where only minimal features are needed to train the classifier, including sentiment-bearing Co-Occurring Terms (COTs) and negation words. This approach yields close to state-of-the-art results. Contrary to results in other domains, the use of negations as features does not have a positive impact in the evaluation results. This method is particularly suited for languages that suffer from a lack of resources, such as sentiment lexicons or parsers, and for those systems that need to function in real-time.

The use of smartphones and tablets has increased significantly in the past years and changed the way we consume news. In this paper, we will describe a news stream aggregating system that automatically recognize and disambiguate geo spatial and meaning bearing entities in news text. The system utilizes the entity definitions and associations in WikiData and Geonames to build a semantic representation of the underlying news text. Its mobile user interface allows users to explore news story clusters on an interactive map and retrieve local news streams for any region of the world.

Recommending news articles entails additional requirements to recommender systems. Such requirements include special consumption patterns, fluctuating itemcollections, and highly sparse user profiles. This workshop (NRS'[email protected]) brought together researchers and practitioners around the topics of designing and evaluating novel news recommender systems. Additionally, we offered a challenge allowing participants to evaluate their recommendation algorithms with actual user feedback.

Online news usually describes various events over multiple topics. Some of them may generate great impact and affection on other events, organizations or people. For example, a bankruptcy news about a big company may generate a great impact on other companies. Detecting this kind of impact helps users better to understand the affection of a specified event and its epidemic. Powerful text mining techniques have been developed to help users to detect topic trends of news articles. However, there is a lack of effective analysis tools that analyze and reveal the news impact in an intuitive approach. In this paper, we introduce Impact Wheel, an explorative visual analysis system for topic driven news impact detection. We describe two unique aspects of Impact Wheel, including 1) topic driven impact analysis and 2) interactive rich context visualization. Experiments on performance evaluation show that our proposed approach outperforms the two baseline methods on topic driven impact analysis. In addition, we demonstrate the power of the Impact Wheel system through a case study, which shows the benefits of this work, especially in support of rich topic data analysis.

This paper addresses the problem of extracting, analyzing and synthesizing valuable information from continuous text streams covering financial information. A text mining framework combining elements from information retrieval, information extraction and natural language processing has been implemented. The framework is utilized to extract information regarding key actors in the domain, how they relate to each other, and how these characteristics evolve over time

In information systems engineering, conceptual models are constructed to assess existing information systems and work out requirements for new ones. As these models serve as a means for communication between customers and developers, it is paramount that both parties understand the models, as well as that the models form a proper basis for the subsequent design and implementation of the systems. New CASE environments are now experimenting with formal modeling languages and various techniques for validating conceptual models, though it seems difficult to come up with a technique that handles the linguistic barriers between the parties involved in a satisfactory manner. In this article, we discuss the theoretical basis of an explanation component implemented for the PPP CASE environment. This component integrates other validation techniques and provides a very flexible natural-language interface to complex model information. It describes properties of the modeling language and the conceptual models in terms familiar to users, and the explanations can be combined with graphical model views. When models are executed, it can justify requested inputs and explain computed outputs by relating trace information to properties of the models.