Zum Inhalt springen
Technische Universität München

Practical Course in Summer Semester 2025:

Projects in Recommender Systems [Online] (IN0012, IN2106)

(Ashmi Banerjee, Dr. Wolfgang Wörndl)

News

[23.01.2025] Web page online

Information

  • This Practical Course is for students in Informatics or related Bachelor's or Master's study programs (module IN0012 or IN2106)
    • The course is conducted in English language
    • The course is conduced online via Zoom (no in-person meetings, no need to travel to Garching)
    • The course is conducted in 2 groups (different advisors) with teams of 3 students each
  • No pre-course meeting
  • Prerequisites
    • Proficiency in programming using Python
    • Good understanding of version controlling such as Git
    • Basic data analysis skills and some understanding and interest in recommender systems
  • Additional details about the content, procedure and deliverables will be shared among participants in a Moodle course
  • Requirements for credits
    • Complete, present and document all parts of the solution for your topic
    • Attendance in all (online) meetings (attendance and contribution in the milestone meetings of your group is mandatory, you can optionally joining the milestone meetings of the other group as well)
    • Grading will be based on the quality of the code and solution as presented and documented
  • It is possible to extend and evaluate your project in a Guided Research of the Master's study programs

Registration

(Preliminary) Schedule

  • Wed 23.04.2025, 16:00: Kickoff meeting (all students, both groups) [online via Zoom]
  • Three milestone meetings, each group separately, see below [online via Zoom]
  • Mon 21.07.2025, 14:00: Final presentations (all students, both groups) [online via Zoom]
  • Fri 25.07.2025: Final deadline to complete and submit documentation

Group A (advised by Ashmi)

Milestone meetings - Tentative dates [online via Zoom]

  • Mon 12.05.2025, 13:00
  • Mon 02.06.2025, 13:00
  • Mon 07.07.2025, 13:00

Topic A1: RAG for Tourism Recommender Systems (TRS) [Team: N.N, N.N., N.N.]

Tourism Recommender Systems (TRS) provide personalized travel suggestions using user preferences and contextual data. Retrieval-augmented generation (RAG) models enhance recommendation quality by combining retrieval and generative capabilities. This proposal aims to improve RAG models for TRS by addressing current limitations, optimizing existing systems, and generating synthetic datasets tailored to tourism scenarios. These advancements will create a robust and scalable framework for enhanced personalization and diversity in TRS.

Proposed Work

  1. Improving Existing Systems
    • Enhance the RAG model from our "RecSoGood" paper (referenced below) by addressing bottlenecks in retrieval and generation components.
    • Integrate domain-specific knowledge for better recommendations.
  2. Synthetic Data Generation
    • Develop synthetic datasets simulating user preferences and diverse travel contexts.
    • Improve model generalization and robustness through tailored data

References

  • Banerjee, Ashmi, Adithi Satish, and Wolfgang Wörndl. "Enhancing Tourism Recommender Systems for Sustainable City Trips Using Retrieval-Augmented Generation." arXiv preprint arXiv:2409.18003 (2024).

Topic A2: Gamification in Sustainable TRS [Team: N.N, N.N., N.N.]

This project aims to develop an application that uses gamification techniques to encourage users to make sustainable choices while planning city trips. The application will nudge users towards eco-friendly transportation, accommodation, dining, and activities by integrating game elements such as points, badges, leaderboards, and challenges. The goal is to visualize the impact of different strategies and foster sustainable tourism. It could also be used to recommend places less visited or during nonpeak seasons through a gamified app.

References

  • Xu, Feifei, Jessika Weber, and Dimitrios Buhalis. "Gamification in tourism." Information and Communication Technologies in Tourism 2014: Proceedings of the International Conference in Dublin, Ireland, January 21-24, 2014. Springer International Publishing, 2013.

Topic A3: Personalization in Conversational Recommender Systems for Tourism (C-TRS) [Team: N.N, N.N., N.N.]

Conversational Tourism Recommender Systems (TRS) provide dynamic, personalized recommendations by interacting with users in real-time. Personalization is crucial for these systems to adapt to individual preferences and deliver meaningful suggestions. This proposal aims to enhance personalization in conversational TRS through a comprehensive literature review, developing a small-scale prototype, and, optionally, generating synthetic datasets tailored to conversational scenarios.

Proposed Work

  1. Literature Review
    • Conduct an in-depth review of existing personalization techniques in conversational systems.
    • Identify gaps and opportunities specific to conversational TRS.
  2. Prototype Implementation
    • Develop a small-scale prototype to test and evaluate personalization strategies in a controlled environment.
    • Focus on integrating adaptive dialogue flows and user preference learning.
  3. Synthetic Data Generation (Optional / if need be)
    • Generate synthetic datasets simulating real-world conversational interactions in tourism contexts.
    • Use these datasets to train and benchmark personalization techniques effectively.

References

  • Rana, Arpit, et al. "User experience and the role of personalization in critiquing-based conversational recommendation." ACM Transactions on the Web 18.4 (2024): 1-21.

Topic A4: Accommodation Recommendations using Booking.com open data [Team: N.N, N.N., N.N.]

Accommodation recommendations are critical to Tourism Recommender Systems (TRS), influencing user satisfaction and decision-making. Leveraging Booking.com’s open data, this project aims to improve accommodation recommendation systems by thoroughly analyzing the available datasets, formulating a novel problem statement, and addressing it through a comprehensive literature review and prototype development.

References

Group W (advised by Wolfgang)

Milestone meetings [online via Zoom]

  • Mon 12.05.2025, 16:00
  • Mon 02.06.2025, 16:00
  • Mon 30.06.2025, 16:00

Topic W1: Knowledge Graphs for Tourism Recommendations [Team: N.N, N.N., N.N.]

Knowledge Graphs (KGs) are an important approach to formally represent semantics by describing entities and their relationships. They seem to be particularly useful in complex domains such as Tourism Recommender Systems (TRS). The goal of this project is to first construct a tourism KG, maybe based on data from DBpedia or similar sites, or review existing tourism KGs that can be used as foundation for a recommender system. The task is then to implemented a recommender application on this KG, with the focus on functionalities that exploit the semantic and hierarchical structure of the KG. A simple example is to not only suggest restaurants that fit a given type of cuisines, but also consider places that offer similar cuisines, based on the KG information. The recommender system and its features should be briefly tested as part of a proof-of-concept.

Ref.: Hendrik, Hendrik, Sri Suning Kusumawardani, and Adhistya Erna Permanasari. "Exploring the Landscape of Tourism Knowledge Graphs: A Systematic Literature Review." 2024 9th International Conference on Information Technology and Digital Applications (ICITDA). IEEE, 2024.

Topic W2: Decentralized Point-of-Interest Recommender System [Team: N.N, N.N., N.N.]

Usually, recommender systems are centralized, i.e. user ratings and all other information is stored and managed on centralized servers. However, the idea of a decentralized recommender system has some advantages, especially in mobile scenarios. In this case, users would manage their preferences, ratings, and other personal information on their mobile devices to improve user control and reduce privacy concerns. The information can then be selectively shared with other users and server-based applications to generate personal recommendations locally on the user device. The scenario is a tourist exploring a city and proactively receiving notifications about interesting points-of-interests in the vicinity on her/his smartphone. The focus of this approach is to separate personalization (i.e. management of user preferences and past behavior) from the actual recommendation. The local user model can then be used to generate personalized recommendations on the smartphone, using the global item model that includes ratings and reviews shared by other (possibly anonymized) users. The goal of this project is to implement a prototype application to put this idea into practice. Part of the project is to investigate user interaction with such a system, allowing the users to specify when or how to receive recommendations, and also which parts of their profile to share with services.

Ref.: Cai, Qiqi, et al. "Distributed Recommendation Systems: Survey and Research Directions." ACM Transactions on Information Systems 43.1 (2024): 1-38.

Topic W3: User Control and Strategies for Influencing Algorithms in TRS [Team: N.N, N.N., N.N.]

Modern recommender systems and other machine learning applications are often black-box systems, users - or the systems themselves - do not really know how the recommendations were generated. It may be desirable to allow users some control and influence over the used algorithms which is also a requirement of the EU Digital Services Act (DSA). Examples in tourism are booking sites of hotels, flights or other services that state something like „commission paid or other factors may affect rankings“ but do not explain the details and provide only simple sorting options to influence rankings. The goal of the topic is to investigate how users can better influence ranking and recommendation algorithms in a TRS, both from a user’s perspective (i.e. how the user interface can be developed to effectively support users) but also algorithmic approaches. Users could for example use sliders to change the weights of different methods in a hybrid recommender system, but additional approaches are necessarily to interact with the parameters of complex recommendations algorithms.

Ref.: Bakalov, Fedor, et al. "An approach to controlling user models and personalization effects in recommender systems." Proceedings of the 2013 International Conference on Intelligent User Interfaces. 2013.

Topic W4: Re-finding Personal Information in TRS [Team: N.N, N.N., N.N.]

Recommender systems are mostly recommending “new” items, i.e. items that the user has not seen before. However, personal information management (PIM) deals with finding items that the user already seen, bookmarked or otherwise interact with. The goal in this project is to implement a prototype application that supports information refinding with a recommender system. The scenario could be a tourism recommender system, with the item space consisting of bookmarks, calender entries, stored markers on maps, documents such as tickets and other relevent items. However, the specific application scenario is flexible in this project. One aspect is to not only respond to an explicit user request, but proactively recommend and show information to the user based on the current situation (location, time). Another aspect is to generate and display explanations for the recommended items, because it may not be clear to the user why an item is shown in a certain use case.

Ref.: Sappelli, Maya, Suzan Verberne, and Wessel Kraaij. "Evaluation of context‐aware recommendation systems for information re‐finding." Journal of the Association for Information Science and Technology 68.4 (2017): 895-910.

Contact

To top