This is the preface for the Proceedings of the Fifth Knowledge-Aware and Conversational Recommender Systems Workshop (KaRS 2023), co-located with the 17th ACM RecSys 2023 conference.
as natural. Finally, this edition of KaRS has reflected the
emergence of several significant milestones, including
In this volume, we include the contributions presented neurosymbolic approaches that aim to bridge the gap
at the Fifth Knowledge-aware and Conversational Rec- between machine learning models and semantic machine
ommender Systems Workshop (KaRS 2023), co-located understanding.
with the 17th ACM Conference on Recommender Sys- The discussion about the connections between these
tems (RecSys 2023) [
This workshop provides a meeting hub where re- technologies and their future interconnections. searchers and companies can showcase their work in Knowledge-aware and Conversational Recommender Systems, as well as discover valuable contacts and mean- 2. Background and Goals ingful connections with other people.
Indeed, while a few years ago the utilization of ex- Recommender systems have become ubiquitous in daily ternal knowledge in literature was sporadic rather than life. They are used in various applications, ranging systematic, in recent years, the widespread adoption of from online shopping to music and movie recommenKnowledge Graphs (KGs) as repositories of structured dations. However, these systems have limitations when information has demonstrated their potential to signif- it comes to interacting with human users [10]. While icantly enhance the performance of recommendation data-driven algorithms have been successful in identimodels. Moreover, the advent of ChatGPT has changed fying latent connections among users and items [11], the users’ perception of Conversational Recommender they often miss a fundamental actor in the loop: the endSystems (CRSs): even people who were not accustomed user. Current research is focusing on new challenges to using conversational agents are now more inclined to such as privacy [12], and new paradigms such as federtheir utilization and perceive conversations with chatbots ated learning [13, 14]. The exploitation of the knowledge about the domain of interest of a catalog via automated reasoning and critiquing approaches is a common behavior of a human user, but it is not well codified in recommendation engine behaviors. One way to overcome these limitations is through knowledge-based apFifth Workshop on Knowledge-aware and Conversational Recommender Systems (KaRS), co-located with the co-located with 17th ACM Conference on Recommender Systems (RecSys 2023)
© 2023 Copyright for this paper by its authors. Use permitted under Creative
CPWrEooUrckReshdoinpgs IhStpN:/c1e6u1r3-w-0s.o7r3g CCoEmmUoRns LWiceonsrekAstthribouptionP4r.0oIncteerenadtiionnagl s(CC(CBYE4U.0)R.-WS.org)
proaches [15, 16, 17, 18, 19], which are now gaining more the traditional goal of accuracy [10] and focuses on
imattention due to the Linking Open Data1 initiative and proving the user experience, engagement, and
satisfacthe availability of large knowledge graphs such as DB- tion [42]. This requires a diverse set of skills and expertise
pedia2 and Wikidata3. These approaches provide rec- from fields such as Machine Learning, Human-Computer
ommendation to users exploiting the domain-specific Interaction, Information Retrieval, and Information
Sysknowledge encoded in ontologies or knowledge graphs, tems, among others.
used to encode the relationships between items, users, The KaRS Workshop brings together researchers and
and other relevant entities. The exploitation of such practitioners to share their research and techniques,
indatasets together with their ontologies is at the basis cluding new design technologies, and identify the next
of many approaches to recommendation and challenges key challenges and emerging topics in the field. The
proposed in the last years such as Knowledge Graph goal is to establish an interdisciplinary community with
embeddings [20, 21, 22, 23, 24], hybrid recommenda- a focus on the exploitation of (semi-)structured
knowltion [18, 25, 26], link prediction [27, 28, 23, 29, 30, 31, 32], edge and conversational approaches for recommender
interpretable recommendation [33, 34, 18], and user mod- systems, which can lead to exciting collaboration
opporeling [35]. Successful workshops and conferences in tunities both for academics and industry practitioners.
the last few years (ISWC, RecSys, UMAP, AAAI, ECAI, In summary, the fith edition of KaRS [
Furthermore, content features become crucial when Topics of interests include, but are not limited to: interaction requires it, such as in Conversational Recommender Systems (CRSs) [38]. These systems are charac- • Models and Feature Engineering: Data models terized by a multi-turn dialogue between the user and based on structured knowledge sources (e.g., Linked the system [39]. The term “conversational” in this con- Open Data, Wikidata, BabelNet, etc.), Semantics-aware text does not necessarily mean that the system conducts approaches exploiting the analysis of textual sources dialogues in natural language, as it may allow more con- (e.g., Wikipedia, Social Web, etc.), Knowledge-aware strained modes of user interaction as well [40]. However, user modeling, Methodological aspects (evaluation prothis type of interaction creates new challenges since it tocols, metrics, and datasets), Logic-based modeling blurs the distinction between recommendation and re- of a recommendation process, Knowledge Representatrieval. A CRS should be able to utilize both short- and tion and Automated Reasoning for recommendation long-term preferences and adapt its behavior promptly engines, Deep learning methods to model semantic when user feedback is provided. There are also other pe- features culiarities of CRSs, such as evaluating the systems beyond accuracy metrics [41]. Moreover, the evaluation of CRSs is a sensitive issue also due to the limited availability of datasets [41]. Although research and development into CRSs has been less prominent for some time, recently, the literature on this topic has been growing notably [39]. • Beyond-Accuracy Recommendation Quality: Using knowledge bases and knowledge graphs to increase recommendation quality (e.g., in terms of novelty, diversity, serendipity, or explainability), Explainable Recommender Systems, Knowledge-aware explanations (compliant with the General Data Protection Regulation) 2.1. Objectives The Fifth Knowledge-aware and Conversational Recommender Systems (KaRS) Workshop is not just another academic event focused on the latest algorithms and approaches for recommendation engines. Instead, it aims to spark a new generation of research that goes beyond 1http://linkeddata.org 2https://dbpedia.org 3https://wikidata.org • Online Studies: Knowledge sources for cross-lingual recommendations, Applications of knowledge-aware recommenders (e.g., music or news recommendation, of-mainstream application areas), User studies (e.g., on the user’s perception of knowledge-based recommendations), field studies • Design of a Conversational Agent: Design and implementation methodologies, Dialogue management (end-to-end, dialogue-state-tracker models), UX de- Milano), Dietmar Jannach (University of Klagenfurt), sign, Dialogue protocol design Daniele Malitesta (Polytechnic University of Bari), Alberto Carlo Maria Mancino (Polytechnic University of • User Modeling and Interfaces: Critiquing and user’s Bari), Olga Marino (Universidad de los Andes), Mirko feedback exploitation, Short- and Long-term user pro- Marras (University of Cagliari), David Massimo (Free ifling and modeling, Preference elicitation, Natural lan- University of Bolzano), Giacomo Medda (University of guage, multimodal, and voice-based interfaces, Next- Cagliari), Cataldo Musto (University of Bari), Franco question problem Maria Nardini (ISTI-CNR), Fedelucio Narducci (Poly• Methodological and Theoretical aspects: Evalua- technic University of Bari), Vincenzo Paparella (Polytion and metrics, Datasets, Theoretical aspects of con- technic University of Bari), Rafaele Perego (ISTI-CNR), versational recommender systems Marco Polignano (University of Bari), Claudio Pomo (Polytechnic University of Bari), Erasmo Purificato (Otto von Guericke University Magdeburg), Azzurra 3. Program Ragone (University of Bari), Yongli Ren (RMIT University), Chiara Renso (ISTI-CNR), Federico Siciliano The program of the half-day workshop consisted of: (Sapienza University of Rome), Marko Tkalcic (Free University of Bozen), Markus Zanker (Free University • a session with the presentation of papers on Conver- of Bozen and University of Klagenfurt).
sational Recommender Systems; • a session with the presentation of papers on Large
Language Models for recommendation; • two sessions with the presentation of papers on
Knowledge-aware Recommender Systems.
KaRS 2023 for their thorough reviews and their detailed feedback they gave to the authors. The PC comprised people from diferent countries and spanning various levels of seniority: Nourah AlRossais (King Saud University), Vito Walter Anelli (Polytechnic University of Bari), Marco Angelini (Sapienza University of Rome), Andrea Bacciu (Sapienza University of Rome), Giacomo Balloccu (University of Cagliari), Pierpaolo Basile (University of Bari), Alejandro Bellogìn (Universidad Autonoma de Madrid), Giovanni Maria Biancofiore (Polytechnic University of Bari), Ludovico Boratto (University of Cagliari), Giandomenico Cornacchia (Polytechnic University of Bari), Humberto Corona (Spotify), Marco de Gemmis (University of Bari), Gerard De Melo (Hasso Plattner Institute and University of Potsdam), Amra Delić University of Sarajevo, Davide Di Ruscio (Università degli Studi dell’Aquila), Francesco Maria Donini Università della Tuscia, Fabrizio Falchi (ISTI-CNR), Antonio Ferrara (Polytechnic University of Bari), Maurizio Ferrari Dacrema (Politecnico di ACM International Conference on Information Recommender systems—beyond matrix completion, and Knowledge Management, CIKM 2019, Bei- Communications of the ACM 59 (2016) 94–102. jing, China, November 3-7, 2019, ACM, 2019, pp. [11] V. W. Anelli, A. Bellogín, T. D. Noia, C. Pomo, Reen3001–3002. visioning the comparison between neural collabo[5] V. W. Anelli, T. D. Noia (Eds.), Proceedings of the rative filtering and matrix factorization, in: RecSys, Second Workshop on Knowledge-aware and Con- ACM, 2021, pp. 521–529. versational Recommender Systems, co-located with [12] V. W. Anelli, L. Belli, Y. Deldjoo, T. D. Noia, A. Fer28th ACM International Conference on Information rara, F. Narducci, C. Pomo, Pursuing privacy in and Knowledge Management, KaRS@CIKM 2019, recommender systems: the view of users and reBeijing, China, November 7, 2019, volume 2601 of searchers from regulations to applications, in: RecCEUR Workshop Proceedings, CEUR-WS.org, 2020. Sys, ACM, 2021, pp. 838–841.
URL: http://ceur-ws.org/Vol-2601. [13] V. W. Anelli, Y. Deldjoo, T. Di Noia, A. Ferrara, [6] V. W. Anelli, P. Basile, T. D. Noia, F. M. Donini, F. Narducci, Federank: User controlled feedback C. Musto, F. Narducci, M. Zanker, Third knowledge- with federated recommender systems, in: D. Hiemaware and conversational recommender systems stra, M.-F. Moens, J. Mothe, R. Perego, M. Potthast, workshop (kars), in: H. J. C. Pampín, M. A. Lar- F. Sebastiani (Eds.), Advances in Information Reson, M. C. Willemsen, J. A. Konstan, J. J. McAuley, trieval, Springer International Publishing, Cham, J. Garcia-Gathright, B. Huurnink, E. Oldridge (Eds.), 2021, pp. 32–47.
RecSys ’21: Fifteenth ACM Conference on Recom- [14] V. W. Anelli, Y. Deldjoo, T. D. Noia, A. Ferrara, Priormender Systems, Amsterdam, The Netherlands, 27 itized multi-criteria federated learning, Intelligenza September 2021 - 1 October 2021, ACM, 2021, pp. Artificiale 14 (2020) 183–200. URL: https://doi.org/ 806–809. 10.3233/IA-200054. doi:10.3233/IA-200054. [7] V. W. Anelli, P. Basile, T. D. Noia, F. M. Donini, [15] R. Burke, Knowledge-based recommender systems.
C. Musto, F. Narducci, M. Zanker, H. Abdollahpouri, encyclopedia of library and information systems: T. Bogers, B. Mobasher, C. Petersen, M. S. Pera Vol. 6.(supplement 32), 2000. (Eds.), Joint Workshop Proceedings of the 3rd Edi- [16] A. Felfernig, G. Friedrich, D. Jannach, M. Zanker, tion of Knowledge-aware and Conversational Rec- Constraint-based recommender systems, in: Recommender Systems (KaRS) and the 5th Edition of ommender Systems Handbook, 2015, pp. 161–190. Recommendation in Complex Environments (Com- [17] M. Zanker, M. Jessenitschnig, W. Schmid, PreferplexRec) co-located with 15th ACM Conference ence reasoning with soft constraints in constrainton Recommender Systems (RecSys 2021), Virtual based recommender systems, Constraints 15 (2010) Event, Amsterdam, The Netherlands, September 25, 574–595. 2021, volume 2960 of CEUR Workshop Proceedings, [18] V. W. Anelli, T. D. Noia, E. D. Sciascio, A. Ragone, CEUR-WS.org, 2021. J. Trotta, How to make latent factors inter[8] V. W. Anelli, P. Basile, G. de Melo, F. M. Donini, pretable by feeding factorization machines with A. Ferrara, C. Musto, F. Narducci, A. Ragone, knowledge graphs, in: C. Ghidini, O. Hartig, M. Zanker, Fourth knowledge-aware and conver- M. Maleshkova, V. Svátek, I. F. Cruz, A. Hogan, sational recommender systems workshop (kars), J. Song, M. Lefrançois, F. Gandon (Eds.), The Semanin: J. Golbeck, F. M. Harper, V. Murdock, M. D. tic Web - ISWC 2019 - 18th International Semantic Ekstrand, B. Shapira, J. Basilico, K. T. Lundgaard, Web Conference, Auckland, New Zealand, October E. Oldridge (Eds.), RecSys ’22: Sixteenth ACM 26-30, 2019, Proceedings, Part I, volume 11778 of Conference on Recommender Systems, Seattle, Lecture Notes in Computer Science, Springer, 2019, WA, USA, September 18 - 23, 2022, ACM, 2022, pp. 38–56. pp. 663–666. URL: https://doi.org/10.1145/3523227. [19] V. W. Anelli, T. D. Noia, E. D. Sciascio, A. Ferrara, 3547412. doi:10.1145/3523227.3547412. A. C. M. Mancino, Sparse feature factorization for [9] V. W. Anelli, P. Basile, G. de Melo, F. M. Donini, recommender systems with knowledge graphs, in: A. Ferrara, C. Musto, F. Narducci, A. Ragone, RecSys, ACM, 2021, pp. 154–165.
M. Zanker (Eds.), Proceedings of the Fourth [20] E. Palumbo, D. Monti, G. Rizzo, R. Troncy, E. BarKnowledge-aware and Conversational Recom- alis, entity2rec: Property-specific knowledge graph mender Systems Workshop co-located with 16th embeddings for item recommendation, Expert Syst. ACM Conference on Recommender Systems (Rec- Appl. 151 (2020) 113235.
Sys 2022), Seattle, WA, USA, September 22nd, 2022, [21] Y. Zhang, X. Xu, H. Zhou, Y. Zhang, Distilling volume 3294 of CEUR Workshop Proceedings, CEUR- structured knowledge into embeddings for explainWS.org, 2022. URL: https://ceur-ws.org/Vol-3294. able and accurate recommendation, in: J. Caverlee, [10] D. Jannach, P. Resnick, A. Tuzhilin, M. Zanker, X. B. Hu, M. Lalmas, W. Wang (Eds.), WSDM ’20: The Thirteenth ACM International Conference on tion in item related domains via a co-factorization Web Search and Data Mining, Houston, TX, USA, model, in: A. Gangemi, R. Navigli, M. Vidal, P. HitFebruary 3-7, 2020, ACM, 2020, pp. 735–743. zler, R. Troncy, L. Hollink, A. Tordai, M. Alam (Eds.), [22] C. Ni, K. S. Liu, N. Torzec, Layered graph embedding The Semantic Web - 15th International Conference, for entity recommendation using wikipedia in the ESWC 2018, Heraklion, Crete, Greece, June 3-7, yahoo! knowledge graph, in: A. E. F. Seghrouchni, 2018, Proceedings, volume 10843 of Lecture Notes G. Sukthankar, T. Liu, M. van Steen (Eds.), Com- in Computer Science, Springer, 2018, pp. 496–511. panion of The 2020 Web Conference 2020, Taipei, [30] M. S. Schlichtkrull, T. N. Kipf, P. Bloem, R. van den Taiwan, April 20-24, 2020, ACM / IW3C2, 2020, pp. Berg, I. Titov, M. Welling, Modeling relational 811–818. data with graph convolutional networks, in: [23] T. Dettmers, P. Minervini, P. Stenetorp, S. Riedel, A. Gangemi, R. Navigli, M. Vidal, P. Hitzler, Convolutional 2d knowledge graph embeddings, in: R. Troncy, L. Hollink, A. Tordai, M. Alam (Eds.), S. A. McIlraith, K. Q. Weinberger (Eds.), Proceedings The Semantic Web - 15th International Conference, of the Thirty-Second AAAI Conference on Artificial ESWC 2018, Heraklion, Crete, Greece, June 3-7, Intelligence, (AAAI-18), the 30th innovative Appli- 2018, Proceedings, volume 10843 of Lecture Notes cations of Artificial Intelligence (IAAI-18), and the in Computer Science, Springer, 2018, pp. 593–607. 8th AAAI Symposium on Educational Advances [31] C. Shang, Y. Tang, J. Huang, J. Bi, X. He, B. Zhou, in Artificial Intelligence (EAAI-18), New Orleans, End-to-end structure-aware convolutional netLouisiana, USA, February 2-7, 2018, AAAI Press, works for knowledge base completion, in: The 2018, pp. 1811–1818. Thirty-Third AAAI Conference on Artificial Intel[24] M. Nayyeri, S. Vahdati, X. Zhou, H. S. Yazdi, ligence, AAAI 2019, The Thirty-First Innovative J. Lehmann, Embedding-based recommendations Applications of Artificial Intelligence Conference, on scholarly knowledge graphs, in: A. Harth, S. Kir- IAAI 2019, The Ninth AAAI Symposium on Edurane, A. N. Ngomo, H. Paulheim, A. Rula, A. L. Gen- cational Advances in Artificial Intelligence, EAAI tile, P. Haase, M. Cochez (Eds.), The Semantic Web - 2019, Honolulu, Hawaii, USA, January 27 - February 17th International Conference, ESWC 2020, Herak- 1, 2019, AAAI Press, 2019, pp. 3060–3067. lion, Crete, Greece, May 31-June 4, 2020, Proceed- [32] X. Wang, X. He, Y. Cao, M. Liu, T. Chua, KGAT: ings, volume 12123 of Lecture Notes in Computer knowledge graph attention network for recommenScience, Springer, 2020, pp. 255–270. dation, in: A. Teredesai, V. Kumar, Y. Li, R. Ros[25] M. Polignano, C. Musto, M. de Gemmis, P. Lops, ales, E. Terzi, G. Karypis (Eds.), Proceedings of the G. Semeraro, Together is better: Hybrid recommen- 25th ACM SIGKDD International Conference on dations combining graph embeddings and contex- Knowledge Discovery & Data Mining, KDD 2019, tualized word representations, in: RecSys, ACM, Anchorage, AK, USA, August 4-8, 2019, ACM, 2019, 2021, pp. 187–198. pp. 950–958. [26] G. Spillo, C. Musto, M. Polignano, P. Lops, [33] X. Wang, D. Wang, C. Xu, X. He, Y. Cao, T.-S. Chua, M. de Gemmis, G. Semeraro, Combining graph neu- Explainable reasoning over knowledge graphs for ral networks and sentence encoders for knowledge- recommendation, in: Proceedings of the AAAI aware recommendations, in: UMAP, ACM, 2023, Conference on Artificial Intelligence, volume 33, pp. 1–12. 2019, pp. 5329–5336. [27] G. He, J. Li, W. X. Zhao, P. Liu, J. Wen, Mining im- [34] V. W. Anelli, T. D. Noia, E. D. Sciascio, A. Ragone, plicit entity preference from user-item interaction J. Trotta, Semantic interpretation of top-n recomdata for knowledge graph completion via adversar- mendations, IEEE Trans. Knowl. Data Eng. 34 (2022) ial learning, in: Y. Huang, I. King, T. Liu, M. van 2416–2428.
Steen (Eds.), WWW ’20: The Web Conference 2020, [35] H. Wang, F. Zhang, J. Wang, M. Zhao, W. Li, X. Xie, Taipei, Taiwan, April 20-24, 2020, ACM / IW3C2, M. Guo, Exploring high-order user preference on 2020, pp. 740–751. the knowledge graph for recommender systems, [28] Y. Cao, X. Wang, X. He, Z. Hu, T. Chua, Unifying ACM Trans. Inf. Syst. 37 (2019) 32:1–32:26. knowledge graph learning and recommendation: [36] S. Badreddine, A. S. d’Avila Garcez, L. Serafini, Towards a better understanding of user preferences, M. Spranger, Logic tensor networks, Artif. Intell. in: L. Liu, R. W. White, A. Mantrach, F. Silvestri, J. J. 303 (2022) 103649.
McAuley, R. Baeza-Yates, L. Zia (Eds.), The World [37] T. Carraro, A. Daniele, F. Aiolli, L. Serafini, Logic Wide Web Conference, WWW 2019, San Francisco, tensor networks for top-n recommendation, in: CA, USA, May 13-17, 2019, ACM, 2019, pp. 151–161. NeSy, volume 3212 of CEUR Workshop Proceedings, [29] G. Piao, J. G. Breslin, Transfer learning for item CEUR-WS.org, 2022, pp. 1–14. recommendations and knowledge graph comple- [38] K. Zhou, W. X. Zhao, S. Bian, Y. Zhou, J.-R. Wen, J. Yu, Improving conversational recommender systems via knowledge graph based semantic fusion, in: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data
Mining, 2020, pp. 1006–1014. [39] D. Jannach, A. Manzoor, W. Cai, L. Chen, A survey on conversational recommender systems, arXiv preprint arXiv:2004.00646 (2020). [40] A. Iovine, F. Narducci, G. Semeraro, Conversational recommender systems and natural language:: A study through the converse framework, Decision
Support Systems 131 (2020) 113250. [41] S. Zhang, K. Balog, Evaluating conversational recommender systems via user simulation, in: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2020, pp. 1512–1520. [42] M. Zanker, L. Rook, D. Jannach, Measuring the impact of online personalisation: Past, present and future, International Journal of Human-Computer Studies 131 (2019) 160–168.