@phdthesis{ubo_mods_00217204, author = {Donkers, Tim}, title = {Harnessing Latent Space Semantics for Enhanced Interpretability of Recommender Systems in Item Retrieval and Online Communication Dynamics}, year = {2024}, address = {Duisburg, Essen}, keywords = {Recommender Systems, Collaborative Filtering, Machine Learning, Deep Learning, Latent Space Semantics, Interpretability, Transparency, User-centered Design, Online Social Networks, Social Polarization, Filter Bubbles, Echo Chambers, Systems Theory, Simulation, Opinion Dynamics, Agent-based Modeling}, abstract = {Recommender systems, in today’s digital age, have emerged as influential algorithmic curators, significantly shaping content consumption, e-commerce, and public opinion. While the majority of research in this area has been anchored in algorithm development and offline evaluation, user-centered considerations are still conspicuously neglected. In particular, pervasive technologies, while powerful, tend to operate as inscrutable \backslashemph{black boxes}, concealing their inner workings from both developers and end users. By exploring the potential for improving the interpretability of the latent information spaces employed by many recommendation models, this work emphasizes the importance of understanding the structural conditions that form the underlying data base. It highlights the depth of insight that can be gained from the intricate relationships between the entities under consideration, and aims to bridge the current gaps in our understanding of recommender systems. The first goal of this thesis is to advance model-based recommender systems by harnessing the power of latent information spaces. To this end, two novel methods are introduced: Tag-enhanced Matrix Factorization (TagMF) and Aspect-based Transparent Memories (ATM). TagMF extends traditional matrix factorization by intertwining associations between items and tags, thereby indirectly inferring user-tag relationships as well. This not only enriches the user-item preference matrix, improving prediction accuracy and mitigating data sparsity issues, but also increases the degree to which the otherwise latent semantics can be interpreted. On the other hand, ATM leverages user reviews and applies deep learning techniques to provide evidence-backed recommendations. By associating the semantic subtleties within its latent space with concrete user utterances, ATM paves the way for transparent recommender systems that more closely resemble how humans justify their evaluations of items. Together, these approaches, validated by user studies, enable users to interactively navigate and influence the recommendation process, increasing both perceived self-efficacy and recommendation quality. The second objective introduces a methodology grounded in simulation, offering a nuanced lens to comprehend social phenomena pertinent to recommender systems in online social networks. Recognizing the gaps in existing research, this approach underscores the significance of psychological and sociological factors in deciphering the impact of these systems. Traditional offline evaluations, predominantly centered on predicting item ratings or rankings, often bypass the diverse influences on decision-making within recommendation tasks. While user-centric experimental studies have sought alignment between recommendation technology and psychological or demographic attributes, they tend to narrow down phenomena to individual experiences. In environments like social networks, where collective dynamics are crucial, broader effects must be considered to fully grasp the societal implications of recommender systems. To bridge these research voids, our methodology harnesses the rich semantic connections inherent in latent information spaces. It seeks to analyze phenomena such as the polarization of ideological groups by tracing the evolution of these semantics. This perspective allows us to understand how the dynamics between individual users, their social environment, and algorithm-driven content distribution together influence the spread of opinions and the manifestation of particular beliefs within online networks. In summary, this thesis presents an alternative approach to studying recommender systems that emphasizes human factors and latent semantics. It argues that, beyond predicting ratings or ranking alternatives, additional information needs can be satisfied by accessing and disclosing latent semantics inferred by contemporary machine learning technologies, ultimately enabling users to properly evaluate system-side suggestions. In addition, the work proposes a simulation paradigm to reconcile psychological and sociological factors influencing decision-making and communication behavior with variations in recommendation technology. These contributions aim to improve our understanding of the complex relationship between users and recommender systems, ultimately enhancing their functionality and impact.}, note = {Dissertation, Universität Duisburg-Essen, 2024, kumulativ}, doi = {10.17185/duepublico/81491}, url = {https://doi.org/10.17185/duepublico/81491}, language = {en} } @inproceedings{kleemann2023, abstract = {Today’s e-commerce websites often provide many different components, such as filters and conversational product advisors, to help users find relevant items. However, filters and advisors are often presented separately and treated as independent entities so that the previous input is discarded when users switch between them. This leads to memory loads and disruptions during the search process. In addition, the reasoning behind the advisors’ results is often not transparent. To overcome these limitations, we propose a novel approach that exploits a graph structure to create an integrated system that allows a seamless coupling between filters and advisors. The integrated system utilizes the graph to suggest appropriate filter values and items based on the user’s answers in the advisor. Moreover, it determines follow-up questions based on the filter values set by the user. The interface visualizes and explains the relationship between a given answer and its relevant features to achieve increased transparency in the guidance process. We report the results of an empirical user study with 120 participants that compares the integrated system to a system in which the filtering and advisory mechanisms operate separately. The findings indicate that displaying recommendations and explanations directly in the filter component can increase acceptance and trust in the system. Similarly, combining the advisor with the filters along with the displayed explanations leads to significantly higher levels of knowledge about the relevant product features.}, address = {Cham}, author = {Kleemann, Timm and Ziegler, Jürgen}, series = {Lecture Notes in Computer Science}, booktitle = {Human-Computer Interaction – INTERACT 2023 : 19th IFIP TC13 International Conference, York, UK, August 28 – September 1, 2023, Proceedings, Part III}, editor = {Abdelnour Nocera, José and Kristı́n Lárusdóttir, Marta and Petrie, Helen and Piccinno, Antonio and Winckler, Marco}, isbn = {9783031422850}, issn = {0302-9743}, volume = {14144}, doi = {10.1007/978-3-031-42286-7_8}, pages = {137–159}, publisher = {Springer Nature Switzerland}, title = {Blending Conversational Product Advisors and Faceted Filtering in a Graph-Based Approach}, url = {https://doi.org/10.1007/978-3-031-42286-7_8}, note = {10.1007/978-3-031-42286-7_8}, language = {en}, keywords = {Search interfaces; explanations; knowledge graph}, year = {2023}, month = {aug}, day = {25}, month_numeric = {8} } @inproceedings{ubo_mods_00199546, author = {Ma, Yuan and Donkers, Tim and Kleemann, Timm and Ziegler, Jürgen}, editor = {Gwizdka, Jacek and Rieh, Soo Young}, title = {An Instrument for measuring users’ meta-intents}, booktitle = {CHIIR ’23: Proceedings of the 2023 Conference on Human Information Interaction and Retrieval}, year = {2023}, publisher = {ACM}, address = {Washington}, pages = {290–302}, abstract = {We propose the concept of meta-intents which represent high-level user preferences related to the interaction and decision-making in conversational recommender systems (CRS) and present a questionnaire instrument for measuring meta-intents. We conducted a two-stage user study, an exploratory study with 212 participants on Prolific, and a confirmatory study with 394 participants on Prolific. We obtained a reliable and stable meta-intents questionnaire with 22 question items, corresponding to seven latent factors (concepts). These seven factors cover important interaction preferences and are closely related to users’ decision-making process. For example, the factor dialog-initiative reflects whether users prefer to follow the system’s guidance or ask their own questions in a CRS. We conducted statistical analyses of meta-intents in two domains (smartphones and hotels), and a general chatbot scenario. We also investigated the influence of additional factors (demography, decision-making style) on meta-intents through Structural Equation Modeling (SEM). Our results provide preliminary evidence that the proposed meta-intents are domain and demography (gender, age) independent. They can be linked to the general decision-making style and can thus be instrumental in translating general decision-making factors into more concrete design guidance for CRS and their potential personalization. Meta-intents also provide a basis for future analyses of interaction behavior in CRS and the development of a cognitively founded theoretical framework.}, isbn = {979-8-4007-0035-4}, doi = {10.1145/3576840}, url = {https://doi.org/10.1145/3576840.3578317}, language = {en} } @inproceedings{ubo_mods_00191230, author = {Ma, Yuan and Kleemann, Timm and Ziegler, Jürgen}, editor = {}, title = {Psychological User Characteristics and Meta-Intents in a Conversational Product Advisor}, booktitle = {Proceedings of the 9th Joint Workshop on Interfaces and Human Decision Making for Recommender Systems}, series = {CEUR Workshop Proceedings}, year = {2022}, publisher = {}, address = {}, volume = {3222}, pages = {18–32}, keywords = {conversational UI design, interactive behavior analysis, decision making, influence of psychological factors on interaction}, abstract = {We present a study investigating psychological characteristics of users of a GUI-style conversational recommender system in a real-world application case. We collected data of 496 customers of an online shop using a conversational product advisor (CPA), using questionnaire responses concerning decision- making style and a set of meta-intents, a concept we propose to represent high-level user preferences related to the decision process in a CPA. We also analyzed anonymized data on users’ interactions in the CPA. Concerning general decision-making style, we could identify two clusters of users who differ in their scores on scales measuring rational and intuitive decision-making. We found evidence that rationality and intuitiveness scores are differently correlated with the proposed meta-intents such as efficiency orientation, interest in detail, and openness for guidance. Relations with interaction data could be observed between rationality/intuitiveness scores and overall time spent in the CPA. Trying to classify users’ decision style from their interactions, however did not yield positive results. Despite the limitation that only a single CPA was studied in a single domain, our results provide evidence that the proposed meta-intents are linked to the general decision-making style of a user and can thus be instrumental in translating general decision-making factors into more concrete design guidance for CPA and their potential personalization.}, issn = {1613-0073}, url = {http://ceur-ws.org/Vol-3222/paper2.pdf}, language = {en} } @inproceedings{ma2022kars, year = {2022}, title = {Meta-Intents in Conversational Recommender Systems}, abstract = {We present a study investigating the psychological characteristics of users and their conversation-related preferences in a conversational recommender system (CRS). We collected data from 260 participants on Prolific, using questionnaire responses concerning decision-making style, conversation-related feature preferences in the smartphone domain, and a set of meta- intents, a concept we propose to represent high-level user preferences related to the interaction and decision-making in CRS. We investigated the relationship between users’ decision-making style, meta-intents and feature preferences through Structural Equation Modeling. We find that decision-making style has a significant influence on meta-intents as well as on feature preferences, however, meta-intents do not have a mediating effect between these two factors, indicating that meta-intents are independent of item feature preferences and may thus be generalizable, domain-independent concepts. Our results provide evidence that the proposed meta-intents are linked to the general decision-making style of a user and can thus be instrumental in translating general decision-making factors into more concrete design guidance for CRS and their potential personalization. As meta-intents seem to be domain-independent factors, we assume meta-intents do not affect users’ various interests in concrete product features and mainly reflect users’ general decision-support needs and interaction preferences in CRS.}, booktitle = {Proceedings of the 4th Edition of Knowledge-aware and Conversational Recommender Systems (KaRS) Workshop @ RecSys 2022}, author = {Yuan, Ma and Donkers, Tim and Kleemann, Timm and Jürgen, Ziegler} } @inproceedings{ubo_mods_00185435, author = {Kleemann, Timm and Loepp, Benedikt and Ziegler, Jürgen}, publisher = {ACM}, address = {New York, NY, USA}, title = {Towards Multi-Method Support for Product Search and Recommending}, booktitle = {Adjunct Proceedings of the 30th ACM Conference on User Modeling, Adaptation and Personalization (UMAP ’22)}, year = {2022}, pages = {74–79}, keywords = {Decision aids; Faceted filtering; Conversational recommender systems; Product advisors; Chatbots; Recommender systems; Search interfaces}, isbn = {978-1-4503-9232-7}, doi = {10.1145/3511047}, url = {https://dl.acm.org/doi/10.1145/3511047.3536408?cid=87958660357}, abstract = {Today, online shops offer a variety of components to support users in finding suitable items, ranging from filters and recommendations to conversational advisors and natural language chatbots. All these methods differ in terms of cognitive load and interaction effort, and, in particular, in their suitability for the specific user. However, it is often difficult for users to determine which method to use to reach their goal. Moreover, as the settings are not propagated between the methods, there is a lack of support for switching components. In this paper, we study the reasons for using the different components in more detail and present an initial proposal for a multi-method approach that provides a more seamless experience, allowing users to freely and flexibly choose from all available methods at any time.} } @inproceedings{ubo_mods_00167803, author = {Ma, Yuan and Kleemann, Timm and Ziegler, Jürgen}, title = {Mixed-Modality Interaction in Conversational Recommender Systems}, booktitle = {Proceedings of the 8th Joint Workshop on Interfaces and Human Decision Making for Recommender Systems}, series = {CEUR Workshop Proceedings}, year = {2021}, publisher = {}, address = {}, volume = {2948}, pages = {21–37}, keywords = {Conversational Recommender Systems; User Interface; Preference Elicitation; Critique-based Recommendations}, abstract = {Recent advances in natural language processing have made modern chatbots and Conversational Recommender Systems (CRS) increasingly intelligent, enabling them to handle more complex user inputs. Still, the interaction with a CRS is often tedious and error-prone. Especially when using written text as the form of conversation, the interaction is often less efficient in comparison to conventional GUI- style interaction. To keep the flexibility and mixed-initiative style of language-based conversation while leveraging the efficiency and simplicity of interacting through graphical widgets, we investigate the de- sign space of integrating GUI elements into text-based conversations. While simple response buttons have already been used in chatbots, the full range of such mixed-modality interactions has not yet been investigated in existing research. We propose two design dimensions along which integrations can be defined and analyze their applicability for preference elicitation and for critiquing the CRS’s responses at different levels. We report a user study in which we investigated user preferences and perceived usability of different techniques based on video prototypes.}, note = {OA platinum}, issn = {1613-0073}, url = {http://ceur-ws.org/Vol-2948/paper2.pdf}, language = {en} } @inproceedings{ubo_mods_00167689, author = {Kleemann, Timm and Wagner, Magdalena and Loepp, Benedikt and Ziegler, Jürgen}, title = {Modeling User Interaction at the Convergence of Filtering Mechanisms, Recommender Algorithms and Advisory Components}, booktitle = {Mensch Und Computer 2021 – Tagungsband}, year = {2021}, publisher = {ACM}, address = {New York, NY, USA}, pages = {531–543}, keywords = {Human factors; User experience; User modeling; Search interfaces; Recommender systems}, isbn = {978-1-4503-8645-6}, doi = {10.1145/3473856}, url = {https://dl.acm.org/doi/10.1145/3473856.3473859?cid=87958660357}, language = {en}, abstract = {A variety of methods is used nowadays to reduce the complexity of product search on e-commerce platforms, allowing users, for example, to specify exactly the features a product should have, but also, just to follow the recommendations automatically generated by the system. While such decision aids are popular with system providers, research to date has mostly focused on individual methods rather than their combination. To close this gap, we propose to support users in choosing the right method for the current situation. As a first step, we report in this paper a user study with a fictitious online shop in which users were able to flexibly use filter mechanisms, rely on recommendations, or follow the guidance of a dialog-based product advisor. We show that from the analysis of the interaction behavior, a model can be derived that allows predicting which of these decision aids is most useful depending on the user’s situation, and how this is affected by demographics and personality.} } @inproceedings{ubo_mods_00156892, author = {Kleemann, Timm and Ziegler, Jürgen}, title = {Distribution sliders: Visualizing data distributions in range selection sliders}, booktitle = {Conference on "Mensch und Computer"}, series = {ACM International Conference Proceeding Series}, year = {2020}, publisher = {Association for Computing Machinery (ACM)}, address = {New York}, pages = {67–78}, isbn = {9781450375405}, doi = {10.1145/3404983.3405512}, abstract = {Sliders are often used to enable users to easily enter preferences for continuous data. Although efforts have already been made to enrich and improve these interaction tools with additional information and visualizations, only rather basic variants of sliders are commonly used in online shops or databases. However, these sliders often provide users only with very limited information about underlying data.We describe and evaluate three different slider designs, which enrich the tools with information in various ways, enabling users to efficiently explore the space of available items and to choose items in an informed manner. In one of the described slider designs we propose a new approach that integrates item recommendations directly into the slider, enabling users to see suitable items within the selection tool. In two user studies we show that these enhancements, both visualizations and recommendations, are powerful methods to directly support users in their searches.} } @inproceedings{ubo_mods_00148660, author = {Donkers, Tim and Kleemann, Timm and Ziegler, Jürgen}, editor = {Paternò, Fabio and Oliver, Nuria}, title = {Explaining Recommendations by Means of Aspect-Based Transparent Memories}, booktitle = {Proceedings of the 25th International Conference on Intelligent User Interfaces}, year = {2020}, publisher = {The Association for Computing Machinery}, address = {New York, NY}, pages = {166–176}, isbn = {978-1-4503-7118-6}, doi = {10.1145/3377325.3377520}, url = {https://dl.acm.org/doi/pdf/10.1145/3377325.3377520}, abstract = {Recommender Systems have seen substantial progress in terms of algorithmic sophistication recently. Yet, the systems mostly act as black boxes and are limited in their capacity to explain why an item is recommended. In many cases recommendations methods are employed in scenarios where users not only rate items, but also convey their opinion on various relevant aspects, for instance by the means of textual reviews. Such user-generated content can serve as a useful source for deriving explanatory information to increase system intelligibility and, thereby, the user’s understanding. We propose a recommendation and explanation method that exploits the comprehensiveness of textual data to make the underlying criteria and mechanisms that lead to a recommendation more transparent. Concretely, the method incorporates neural memories that store aspect-related opinions extracted from raw review data. We apply attention mechanisms to transparently write and read information from memory slots. Besides customary offline experiments, we conducted an extensive user study. The results indicate that our approach achieves a higher overall quality of explanations compared to a state-of-the-art baseline. Utilizing Structural Equation Modeling, we additionally reveal three linked key factors that constitute explanation quality: Content adequacy, presentation adequacy, and linguistic adequacy.} } @inproceedings{ubo_mods_00142455, author = {Loepp, Benedikt and Donkers, Tim and Kleemann, Timm and Ziegler, Jürgen}, title = {Impact of Consuming Suggested Items on the Assessment of Recommendations in User Studies on Recommender Systems}, booktitle = {Proceedings of the 28th International Joint Conference on Artificial Intelligence (IJCAI ’19)}, year = {2019}, publisher = {IJCAI Organization}, pages = {6201–6205}, keywords = {Recommender Systems}, doi = {10.24963/ijcai.2019/863}, url = {https://doi.org/10.24963/ijcai.2019/863}, abstract = {User studies are increasingly considered important in research on recommender systems. Although participants typically cannot consume any of the recommended items, they are often asked to assess the quality of recommendations and of other aspects related to user experience by means of questionnaires. Not being able to listen to recommended songs or to watch suggested movies, might however limit the validity of the obtained results. Consequently, we have investigated the effect of consuming suggested items. In two user studies conducted in different domains, we showed that consumption may lead to differences in the assessment of recommendations and in questionnaire answers. Apparently, adequately measuring user experience is in some cases not possible without allowing users to consume items. On the other hand, participants sometimes seem to approximate the actual value of recommendations reasonably well depending on domain and provided information.} } @inproceedings{ubo_mods_00139552, author = {Kleemann, Timm and Ziegler, Jürgen}, title = {Integration of Dialog-based Product Advisors into Filter Systems}, booktitle = {Proceedings of the Conference on Mensch und Computer}, series = {ACM International Conference Proceeding Series}, year = {2019}, publisher = {ACM Press}, address = {New York}, pages = {67–77}, keywords = {Dialogbasierte Produktberater, Filtersysteme}, isbn = {978-1-4503-7198-8}, doi = {10.1145/3340764.3340786}, abstract = { Different techniques such as search functions or recommendation components are used today to support the often complex product search on the Internet. Faceted filter systems that successively limit the result set according to the set filter settings have proven to be quite successful. However, this method requires clear objectives and domain knowledge on the part of the users. As an alternative, conversational product advisors who select suitable products on the basis of a sequence of questions have gained more importance in recent times, whereby the questions are based more on the tasks and application scenarios of the users than on the technical properties of the products. However, there is currently a lack of approaches that integrate filter systems and conversational advisors in a meaningful and closely coupled way. In this paper an integrated approach is presented, where users can switch between filter systems and advisory dialogues, whereby selection actions in one component have a consistent and transparent effect on the other component and can be further adjusted there. The aim is to better support users with different levels of knowledge of the product type concerned. We describe the requirements for such integrated systems resulting from our approach and report on a user study in which the user behavior and the subjective evaluation were examined in a prototypical implementation.} } @article{ubo_mods_00109856, author = {Loepp, Benedikt and Donkers, Tim and Kleemann, Timm and Ziegler, Jürgen}, volume = {121}, pages = {21–41}, title = {Interactive Recommending with Tag-Enhanced Matrix Factorization (TagMF)}, journal = {International Journal of Human Computer Studies}, year = {2019}, keywords = {Collaborative Filtering, Empirical studies, Human factors, Interactive recommending, Matrix Factorization, Recommender Systems, Tags, User experience, User interfaces, User profiles}, doi = {10.1016/j.ijhcs.2018.05.002}, url = {https://doi.org/10.1016/j.ijhcs.2018.05.002}, abstract = {We introduce TagMF, a model-based Collaborative Filtering method that aims at increasing transparency and offering richer interaction possibilities in current Recommender Systems. Model-based Collaborative Filtering is currently the most popular method that predominantly uses Matrix Factorization: This technique achieves high accuracy in recommending interesting items to individual users by learning latent factors from implicit feedback or ratings the community of users provided for the items. However, the model learned and the resulting recommendations can neither be explained, nor can users be enabled to influence the recommendation process except by rating (more) items. In TagMF, we enhance a latent factor model with additional content information, specifically tags users provided for the items. The main contributions of our method are to use this integrated model to elucidate the hidden semantics of the latent factors and to let users interactively control recommendations by changing the influence of the factors through easily comprehensible tags: Users can express their interests, interactively manipulate results, and critique recommended items—at cold-start when no historical data is yet available for a new user, as well as in case a long-term profile representing the current user’s preferences already exists. To validate our method, we performed offline experiments and conducted two empirical user studies where we compared a recommender that employs TagMF against two established baselines, standard Matrix Factorization based on ratings, and a purely tag-based interactive approach. This user-centric evaluation confirmed that enhancing a model-based method with additional information positively affects perceived recommendation quality. Moreover, recommendations were considered more transparent and users were more satisfied with their final choice. Overall, learning an integrated model and implementing the interactive features that become possible as an extension to contemporary systems with TagMF appears beneficial for the subjective assessment of several system aspects, the level of control users are able to exert over the recommendation process, as well as user experience in general.} } @inproceedings{ubo_mods_00116566, author = {Loepp, Benedikt and Donkers, Tim and Kleemann, Timm and Ziegler, Jürgen}, title = {Impact of Item Consumption on Assessment of Recommendations in User Studies}, booktitle = {Proceedings of the 12th ACM Conference on Recommender Systems (RecSys ’18)}, year = {2018}, publisher = {ACM}, address = {New York, NY, USA}, pages = {49–53}, keywords = {User Studies}, isbn = {978-1-4503-5901-6}, doi = {10.1145/3240323.3240375}, url = {https://dl.acm.org/doi/10.1145/3240323.3240375?cid=87958660357}, abstract = {In user studies of recommender systems, participants typically cannot consume the recommended items. Still, they are asked to assess recommendation quality and other aspects related to user experience by means of questionnaires. Without having listened to recommended songs or watched suggested movies, however, this might be an error-prone task, possibly limiting validity of results obtained in these studies. In this paper, we investigate the effect of actually consuming the recommended items. We present two user studies conducted in different domains showing that in some cases, differences in the assessment of recommendations and in questionnaire results occur. Apparently, it is not always possible to adequately measure user experience without allowing users to consume items. On the other hand, depending on domain and provided information, participants sometimes seem to approximate the actual value of recommendations reasonably well.} } @inproceedings{ubo_mods_00116218, author = {Angenendt, Kristina and Bormann, Jeanny and Donkers, Tim and Goebel, Tabitha and Kizina, Anna and Kleemann, Timm and Michael, Lisa and Raja, Hifsa and Sachs, Franziska and Schneegass, Christina and Sinzig, Lisa-Maria and Steffen, Juliane and Manske, Sven and Hecking, Tobias and Hoppe, Ulrich Heinz}, editor = {Rathmayer, Sabine and Pongratz, Hans}, title = {ConceptCloud -Entwicklung einer Applikation zur Unterstützung von Reflexionsprozessen im Online-Lernportal Go-Lab}, booktitle = {Proceedings of DeLFI Workshops 2015}, series = {CEUR Workshop Proceedings}, year = {2015}, publisher = {CEUR-WS}, address = {Aachen}, volume = {1443}, pages = {132–135}, issn = {1613-0073}, url = {http://ceur-ws.org/Vol-1443/paper26.pdf} }