Analyzing sequences of interactions between users and items, sequential recommendation models can learn user intent and make predictions about the next item. Next to item interactions, most systems also have interactions with what we call non-item pages: these pages are not related to specific items but still can provide insights into the user’s interests, as, for example, navigation pages. We therefore propose a general way to include these non-item pages in sequential recommendation models to enhance next-item prediction. First, we demonstrate the influence of non-item pages on following interactions using the hypotheses testing framework HypTrails and propose methods for representing non-item pages in sequential recommendation models. Subsequently, we adapt popular sequential recommender models to integrate non-item pages and investigate their performance with different item representation strategies as well as their ability to handle noisy data. To show the general capabilities of the models to integrate non-item pages, we create a synthetic dataset for a controlled setting and then evaluate the improvements from including non-item pages on two real-world datasets. Our results show that non-item pages are a valuable source of information, and incorporating them in sequential recommendation models increases the performance of next-item prediction across all analyzed model architectures.

The ever-growing corpus of scientific literature presents significant challenges for researchers with respect to discovery, management, and annotation of relevant publications. Traditional platforms like Semantic Scholar, BibSonomy, and Zotero offer tools for literature management, but largely require manual laborious and error-prone input of tags and metadata. Here, we introduce a novel retrieval augmented generation system that leverages chat-based large language models (LLMs) to streamline and enhance the process of publication management. It provides a unified chat-based interface, enabling intuitive interactions with various backends, including Semantic Scholar, BibSonomy, and the Zotero Webscraper. It supports two main use-cases: (1) Explorative Search & Retrieval - leveraging LLMs to search for and retrieve both specific and general scientific publications, while addressing the challenges of content hallucination and data obsolescence; and (2) Cataloguing & Management - aiding in the organization of personal publication libraries, in this case BibSonomy, by automating the addition of metadata and tags, while facilitating manual edits and updates. We compare our system to different LLM models in three different settings, including a user study, and we can show its advantages in different metrics.

Abstract The term Behavioral Networks describes networks that contain relational information on human behavior. This ranges from social networks that contain friendships or cooperations between individuals, to navigational networks that contain geographical or web navigation, and many more. Understanding the forces driving behavior within these networks can be beneficial to improving the underlying network, for example, by generating new hyperlinks on websites, or by proposing new connections and friends on social networks. Previous approaches considered different hypotheses on a single network and evaluated which hypothesis fits best. These hypotheses can represent human intuition and expert opinions or be based on previous insights. In this work, we extend these approaches to enable the comparison of a single hypothesis between multiple networks. We unveil several issues of naive approaches that potentially impact comparisons and lead to undesired results. Based on these findings, we propose a framework with five flexible components that allow addressing specific analysis goals tailored to the application scenario. We show the benefits and limits of our approach by applying it to synthetic data and several real-world datasets, including web navigation, bibliometric navigation, and geographic navigation. Our work supports practitioners and researchers with the aim of understanding similarities and differences in human behavior between environments.

Processing symbolic mathematics algorithmically is an important field of research. It has applications in computer algebra systems and supports researchers as well as applied mathematicians in their daily work. Recently, exploring the ability of neural networks to grasp mathematical concepts has received special attention. One complex task for neural networks is to understand the relation of two mathematical expressions to each other. Despite the advances in learning mathematical relationships, previous studies are limited by small-scale datasets, relatively simple formula construction by few axiomatic rules and even artifacts in the data. With this work, we aim at overcoming these limitations and provide a deeper insight into the representation power of neural networks for classifying mathematical relations. We introduce a novel data generation algorithm to allow for more complex formula compositions and fully include mathematical fields up to high-school level. We research several tree-based and sequential neural architectures for classifying mathematical relations and conduct a systematic analysis of the models against rule-based as well as neural baselines with a focus on varying dataset complexity, generalization abilities, and understanding of syntactical patterns. Our findings show the potential of deep learning models to distinguish high-school level mathematical concepts.

Abstract Climate models are the tool of choice for scientists researching climate change. Like all models they suffer from errors, particularly systematic and location-specific representation errors. One way to reduce these errors is model output statistics (MOS) where the model output is fitted to observational data with machine learning. In this work, we assess the use of convolutional Deep Learning climate MOS approaches and present the ConvMOS architecture which is specifically designed based on the observation that there are systematic and location-specific errors in the precipitation estimates of climate models. We apply ConvMOS models to the simulated precipitation of the regional climate model REMO, showing that a combination of per-location model parameters for reducing location-specific errors and global model parameters for reducing systematic errors is indeed beneficial for MOS performance. We find that ConvMOS models can reduce errors considerably and perform significantly better than three commonly used MOS approaches and plain ResNet and U-Net models in most cases. Our results show that non-linear MOS models underestimate the number of extreme precipitation events, which we alleviate by training models specialized towards extreme precipitation events with the imbalanced regression method DenseLoss. While we consider climate MOS, we argue that aspects of ConvMOS may also be beneficial in other domains with geospatial data, such as air pollution modeling or weather forecasts.

At the present time, sequential item recommendation models are compared by calculating metrics on a small item subset (target set) to speed up computation. The target set contains the relevant item and a set of negative items that are sampled from the full item set. Two well-known strategies to sample negative items are uniform random sampling and sampling by popularity to better approximate the item frequency distribution in the dataset. Most recently published papers on sequential item recommendation rely on sampling by popularity to compare the evaluated models. However, recent work has already shown that an evaluation with uniform random sampling may not be consistent with the full ranking, that is, the model ranking obtained by evaluating a metric using the full item set as target set, which raises the question whether the ranking obtained by sampling by popularity is equal to the full ranking. In this work, we re-evaluate current state-of-the-art sequential recommender models from the point of view, whether these sampling strategies have an impact on the final ranking of the models. We therefore train four recently proposed sequential recommendation models on five widely known datasets. For each dataset and model, we employ three evaluation strategies. First, we compute the full model ranking. Then we evaluate all models on a target set sampled by the two different sampling strategies, uniform random sampling and sampling by popularity with the commonly used target set size of 100, compute the model ranking for each strategy and compare them with each other. Additionally, we vary the size of the sampled target set. Overall, we find that both sampling strategies can produce inconsistent rankings compared with the full ranking of the models. Furthermore, both sampling by popularity and uniform random sampling do not consistently produce the same ranking when compared over different sample sizes. Our results suggest that like uniform random sampling, rankings obtained by sampling by popularity do not equal the full ranking of recommender models and therefore both should be avoided in favor of the full ranking when establishing state-of-the-art.

When users interact with the Web today, they leave sequential digital trails on a massive scale. Examples of such human trails include Web navigation, sequences of online restaurant reviews, or online music play lists. Understanding the factors that drive the production of these trails can be useful, for example, for improving underlying network structures, predicting user clicks, or enhancing recommendations. In this work, we present a method called HypTrails for comparing a set of hypotheses about human trails on the Web, where hypotheses represent beliefs about transitions between states. Our method utilizes Markov chain models with Bayesian inference. The main idea is to incorporate hypotheses as informative Dirichlet priors and to calculate the evidence of the data under them. For eliciting Dirichlet priors from hypotheses, we present an adaption of the so-called (trial) roulette method, and to compare the relative plausibility of hypotheses, we employ Bayes factors. We demonstrate the general mechanics and applicability of HypTrails by performing experiments with (i) synthetic trails for which we control the mechanisms that have produced them and (ii) empirical trails stemming from different domains including Web site navigation, business reviews, and online music played. Our work expands the repertoire of methods available for studying human trails.

Social tagging systems have established themselves as a quick and easy way to organize information by annotating resources with tags. In recent work, user behavior in social tagging systems was studied, that is, how users assign tags, and consume content. However, it is still unclear how users make use of the navigation options they are given. Understanding their behavior and differences in behavior of different user groups is an important step towards assessing the effectiveness of a navigational concept and improving it to better suit the users' needs. In this work, we investigate navigation trails in the popular scholarly social tagging system BibSonomy from six years of log data. We discuss dynamic browsing behavior of the general user population and show that different navigational subgroups exhibit different navigational traits. Furthermore, we provide strong evidence that the semantic nature of the underlying folksonomy is an essential factor for explaining navigation.

We present a new method for detecting interpretable subgroups with exceptional transition behavior in sequential data. Identifying such patterns has many potential applications, e.g., for studying human mobility or analyzing the behavior of internet users. To tackle this task, we employ exceptional model mining, which is a general approach for identifying interpretable data subsets that exhibit unusual interactions between a set of target attributes with respect to a certain model class. Although exceptional model mining provides a well-suited framework for our problem, previously investigated model classes cannot capture transition behavior. To that end, we introduce first-order Markov chains as a novel model class for exceptional model mining and present a new interestingness measure that quantifies the exceptionality of transition subgroups. The measure compares the distance between the Markov transition matrix of a subgroup and the respective matrix of the entire data with the distance of random dataset samples. In addition, our method can be adapted to find subgroups that match or contradict given transition hypotheses. We demonstrate that our method is consistently able to recover subgroups with exceptional transition models from synthetic data and illustrate its potential in two application examples. Our work is relevant for researchers and practitioners interested in detecting exceptional transition behavior in sequential data.

Social tagging systems have established themselves as an important part in today’s Web and have attracted the interest of our research community in a variety of investigations. Henceforth, several aspects of social tagging systems have been discussed and assumptions have emerged on which our community builds their work. Yet, testing such assumptions has been difficult due to the absence of suitable usage data in the past. In this work, we thoroughly investigate and evaluate four aspects about tagging systems, covering social interaction, retrieval of posted resources, the importance of the three different types of entities, users, resources, and tags, as well as connections between these entities’ popularity in posted and in requested content. For that purpose, we examine live server log data gathered from the real-world, public social tagging system BibSonomy. Our empirical results paint a mixed picture about the four aspects. Although typical assumptions hold to a certain extent for some, other aspects need to be reflected in a very critical light. Our observations have implications for the understanding of social tagging systems and the way they are used on the Web. We make the dataset used in this work available to other researchers.

When users interact with the Web today, they leave sequential digital trails on a massive scale. Examples of such human trails include Web navigation, sequences of online restaurant reviews, or online music play lists. Understanding the factors that drive the production of these trails can be useful for e.g., improving underlying network structures, predicting user clicks or enhancing recommendations. In this work, we present a general approach called HypTrails for comparing a set of hypotheses about human trails on the Web, where hypotheses represent beliefs about transitions between states. Our approach utilizes Markov chain models with Bayesian inference. The main idea is to incorporate hypotheses as informative Dirichlet priors and to leverage the sensitivity of Bayes factors on the prior for comparing hypotheses with each other. For eliciting Dirichlet priors from hypotheses, we present an adaption of the so-called (trial) roulette method. We demonstrate the general mechanics and applicability of HypTrails by performing experiments with (i) synthetic trails for which we control the mechanisms that have produced them and (ii) empirical trails stemming from different domains including website navigation, business reviews and online music played. Our work expands the repertoire of methods available for studying human trails on the Web.

The emergence of what is called the social web and the continuing stream of new applications and community-based platforms including Facebook, Twitter, LinkedIn and others had a substantial impact on recommender systems research and practice over the last years in different ways.

No abstract available.

In this article, the authors present a novel approach for computing semantic relatedness and conduct a large-scale study of it on Wikipedia. Unlike existing semantic analysis methods that utilize Wikipedia’s content or link structure, the authors propose to use human navigational paths on Wikipedia for this task. The authors obtain 1.8 million human navigational paths from a semi-controlled navigation experiment – a Wikipedia-based navigation game, in which users are required to find short paths between two articles in a given Wikipedia article network. The authors’ results are intriguing: They suggest that (i) semantic relatedness computed from human navigational paths may be more precise than semantic relatedness computed from Wikipedia’s plain link structure alone and (ii) that not all navigational paths are equally useful. Intelligent selection based on path characteristics can improve accuracy. The authors’ work makes an argument for expanding the existing arsenal of data sources for calculating semantic relatedness and to consider the utility of human navigational paths for this task.

No abstract available.

Recommender System Challenges such as the Netflix Prize, KDD Cup, etc. have contributed vastly to the development and adoptability of recommender systems. Each year a number of challenges or contests are organized covering different aspects of recommendation. In this tutorial and panel, we present some of the factors involved in successfully organizing a challenge, whether for reasons purely related to research, industrial challenges, or to widen the scope of recommender systems applications.

The exponential growth of the social web poses challenges and new opportunities for recommender systems. The social web has turned information consumers into active contributors creating massive amounts of information. Finding relevant and interesting content at the right time and in the right context is challenging for existing recommender approaches. At the same time, social systems by their definition encourage interaction between users and both online content and other users, thus generating new sources of knowledge for recommender systems. Web 2.0 users explicitly provide personal information and implicitly express preferences through their interactions with others and the system (e.g. commenting, friending, rating, etc.). These various new sources of knowledge can be leveraged to improve recommendation techniques and develop new strategies which focus on social recommendation. The Social Web provides huge opportunities for recommender technology and in turn recommender technologies can play a part in fuelling the success of the Social Web phenomenon.

Contests and challenges have energized researchers and focused attention in many fields recently, including recommender systems. At the 2008 RecSys conference, winners were announced for a contest proposing new startup companies. The 2009 conference featured a panel reflecting on the then recently completed Netflix challenge.

Recent research provides evidence for the presence of emergent semantics in collaborative tagging systems. While several methods have been proposed, little is known about the factors that influence the evolution of semantic structures in these systems. A natural hypothesis is that the quality of the emergent semantics depends on the pragmatics of tagging: Users with certain usage patterns might contribute more to the resulting semantics than others. In this work, we propose several measures which enable a pragmatic differentiation of taggers by their degree of contribution to emerging semantic structures. We distinguish between categorizers, who typically use a small set of tags as a replacement for hierarchical classification schemes, and describers, who are annotating resources with a wealth of freely associated, descriptive keywords. To study our hypothesis, we apply semantic similarity measures to 64 different partitions of a real-world and large-scale folksonomy containing different ratios of categorizers and describers. Our results not only show that "verbose" taggers are most useful for the emergence of tag semantics, but also that a subset containing only 40% of the most 'verbose' taggers can produce results that match and even outperform the semantic precision obtained from the whole dataset. Moreover, the results suggest that there exists a causal link between the pragmatics of tagging and resulting emergent semantics. This work is relevant for designers and analysts of tagging systems interested (i) in fostering the semantic development of their platforms, (ii) in identifying users introducing "semantic noise", and (iii) in learning ontologies.

The challenge to provide tag recommendations for collaborative tagging systems has attracted quite some attention of researchers lately. However, most research focused on the evaluation and development of appropriate methods rather than tackling the practical challenges of how to integrate recommendation methods into real tagging systems, record and evaluate their performance. In this paper we describe the tag recommendation framework we developed for our social bookmark and publication sharing system BibSonomy. With the intention to develop, test, and evaluate recommendation algorithms and supporting cooperation with researchers, we designed the framework to be easily extensible, open for a variety of methods, and usable independent from BibSonomy. Furthermore, this paper presents a first evaluation of two exemplarily deployed recommendation methods.

Social bookmarking systems are becoming increasingly important data sources for bootstrapping and maintaining Semantic Web applications. Their emergent information structures have become known as folksonomies. A key question for harvesting semantics from these systems is how to extend and adapt traditional notions of similarity to folksonomies, and which measures are best suited for applications such as community detection, navigation support, semantic search, user profiling and ontology learning. Here we build an evaluation framework to compare various general folksonomy-based similarity measures, which are derived from several established information-theoretic, statistical, and practical measures. Our framework deals generally and symmetrically with users, tags, and resources. For evaluation purposes we focus on similarity between tags and between resources and consider different methods to aggregate annotations across users. After comparing the ability of several tag similarity measures to predict user-created tag relations, we provide an external grounding by user-validated semantic proxies based on WordNet and the Open Directory Project. We also investigate the issue of scalability. We find that mutual information with distributional micro-aggregation across users yields the highest accuracy, but is not scalable; per-user projection with collaborative aggregation provides the best scalable approach via incremental computations. The results are consistent across resource and tag similarity.

In this paper, we present the foundations for mining frequent tri-concepts, which extend the notion of closed item-sets to three-dimensional data to allow for mining folk-sonomies. We provide a formal definition of the problem, and present an efficient algorithm for its solution as well as experimental results on a large real-world example.

Document representations for text classification are typically based on the classical bag-of-words paradigm. This approach comes with deficiencies that motivate the integration of features on a higher semantic level than single words. In this paper we propose an enhancement of the classical document representation through concepts extracted from background knowledge. Boosting is used for actual classification. Experimental evaluations on two well known text corpora support our approach through consistent improvement of the results.

Text document clustering plays an important role in providing intuitive navigation and browsing mechanisms by organizing large sets of documents into a small number of meaningful clusters. The bag of words representation used for these clustering methods is often unsatisfactory as it ignores relationships between important terms that do not cooccur literally. In order to deal with the problem, we integrate core ontologies as background knowledge into the process of clustering text documents. Our experimental evaluations compare clustering techniques based on pre-categorizations of texts from Reuters newsfeeds and on a smaller domain of an eLearning course about Java. In the experiments, improvements of results by background knowledge compared to a baseline without background knowledge can be shown in many interesting combinations.

Text clustering typically involves clustering in a high dimensional space, which appears difficult with regard to virtually all practical settings. In addition, given a particular clustering result it is typically very hard to come up with a good explanation of why the text clusters have been constructed the way they are. We propose a new approach for applying background knowledge (in terms of an ontology) during preprocessing in order to improve clustering results and allow for selection between results. The results may be distinguished and explained by the corresponding selection of concepts in the ontology. Our results compare favourably with a sophisticated baseline preprocessing strategy.