Markus Jessenitschnig

Case-studies on exploiting explicit customer requirements in recommender systems

Recommender Systems (RS) suggest useful and interesting items to users in order to increase user satisfaction and online conversion rates. They typically exploit explicit or implicit user feedback such as ratings, buying records or clickstream data and apply statistical methods to derive recommendations. This paper focuses on explicitly formulated customer requirements as the sole type of user feedback. Its contribution lies in comparing different techniques such as knowledge- and utility-based methods, collaborative filtering, association rule mining as well as hybrid variants when user models consist solely of explicit customer requirements. We examine how this type of user feedback can be exploited for personalization in e-commerce scenarios. Furthermore, examples of actual online shops are developed where such contextual user information is available, demonstrating how more efficient RS configurations can be implemented. Results indicate that, especially for new users, explicit customer requirements are a useful source of feedback for personalization and hybrid configurations of collaborative and knowledge-based techniques achieve best results.

Comparing Recommendation Strategies in a Commercial Context

From an industrial perspective, recommender systems constitute the base technology for providing interactivity and personalization in electronic business-to-consumer marketplaces. Robin Burke distinguishes between five different recommendation techniques: collaborative, content based, utility based, demographic, and knowledge based.

etPlanner: An IT Framework for Comprehensive and Integrative Travel Guidance

Recommender systems in travel industry make helpful and persuasive product and service suggestions and thus reduce the burden of information overload and domain complexity for users. Within the scope of the etPlanner project we innovate existing technology by considering all trip phases and by supporting next to standard Web interfaces also different mobile devices.

Preference reasoning with soft constraints in constraint-based recommender systems

A recommender system (RS) supports online users in e-commerce by proposing products that are assumed to be both useful and interesting. Knowledge-based recommendation systems form one branch of these online sales support systems that is particularly relevant for high-involvement product domains like consumer electronics, financial services or tourism. A constraint-based RS is a specific variant of a knowledge-based RS that builds on a CSP formalism for problem representation and solving. This article formalizes the different variants of a constraint-based recommendation problem based on consistency and the empirical part compares the performance of different constraint-based recommendation mechanisms in offline experiments on historical data.

Collaborative Feature-Combination Recommender Exploiting Explicit and Implicit User Feedback

Collaborative filtering (CF) is currently the most popular technique used in commercial recommender systems. Algorithms of this type derive personalized product propositions for customers by exploitingstatistics derived from vast amounts of transaction data.Traditionally, basic CF algorithms have exploited a single category of ratings despite the fact that on many platforms a variety of different forms of user feedback are available for personalization and recommendation. In this paper we explore a collaborative feature-combination algorithm that concurrently exploits multiple aspects of the user model like clickstream data, sales transactions and explicit user requirements to overcome some known shortcomingsof CF like the cold-start problem for new users. We validate our contribution by evaluating it against the standard user-to-user CF algorithm using a dataset from a commercial Web shop. Evaluation results indicate considerable improvements in terms of user coverageand accuracy.

Developing a Conversational Travel Advisor with ADVISOR SUITE

Due to the inherent complexity of building highly-interactive and personalized web applications, the development of a web-based travel advisory system can be a costly and time-consuming task. We see this as one of the major obstacles to a more widespread adoption of such systems in particular with respect to small and medium-sized companies and e-Tourism platforms. The goal of the ADVISOR SUITE project discussed in this paper is thus to provide an off-the-shelf framework and development environment that allows us to build intelligent and easy-to-maintain advisory applications in a cost-efficient way: The main pillars of the presented system are therefore an integrated graphical modelling-environment, the provision of different domain-independent recommendation algorithms, as well as model-based mechanisms to fully generate functional web applications based on declarative definitions in a central knowledge repository. The paper discusses the core concepts and main functionalities of the system by means of an example of an interactive travel advisor developed for an Austrian spa resort.

An Automated Approach for Deriving Semantic Annotations of Tourism Products based on Geospatial Information

High quality product data is a necessary prerequisite for supporting efficient browsing and recommendation procedures on e-commerce platforms. This is especially true for the tourism domain where an abundance of information can easily overwhelm users. Although classification data such as to which category (e.g. accommodation, restaurant or sight) a tourism product belongs is usually directly available, qualitative information, such as proximity to a lake or opportunities for dining or shopping, is rarely provided in a structured way. As a consequence, users can not restrict their search on these criteria; rather, it would require costly manual information acquisition efforts. In this paper we propose an approach that automatically associates such qualitative concepts with tourism products based on their geographic coordinates and their spatial proximity. An initial evaluation of the approach that considered automatically generated annotations within different regions suggests that it can be used as an alternative to domain experts.

Automated semantic annotations of tourism resources based on geospatial data.

Web 2.0 applications, now common in the tourism domain, make it easy to share travel related experiences and opinions, thus, leading to the creation of enormous amounts of user generated content. This type of tourist information must, however, be preprocessed by structuring and aggregating it in order to avoid overwhelming users. In this article we, therefore, propose a knowledge-based approach that exploits spatial proximity to annotate resources with qualitative semantic concepts for tourism products, such as proximity to a lake or opportunities for specific sporting activities. Finally, the article describes the successful prototypical implementation of the proposed technique and provides insight into a practical usage scenario

A Generic User Modeling Component for Hybrid Recommendation Strategies

Over the last decade, recommendation systems (RS) have matured into a valuable approach for assisting online customers in navigating through large product or information spaces. The associated research has described and evaluated a variety of different techniques for proposing items of interest to customers. However, each of these techniques also suffers from several shortcomings. Therefore, depending on the application domain and the availability of background knowledge some algorithms and hybrid variants may be more applicable than others. However, most commercial recommendation systems are monolithic in the sense that they support only a limited subset of recommendation techniques.In this paper we therefore present ISeller, a proven industrial-strength recommendation framework for personalizing small to medium-scale e-commerce platforms. ISeller supports all basic recommendation techniques and, due to its modular architecture, hybrid variants as well. This paper focuses in particular on the generic user modeling component of ISeller as it is the prerequisite for supporting different recommendation techniques within the same application infrastructure. Furthermore, we present an application scenario showing the generic nature and wide applicability of the described user modeling component in the domain of map-based recommendations.

A hybrid similarity concept for browsing semi-structured product items

Personalization, information filtering and recommendation are key techniques helping online-customers to orientate themselves in e-commerce environments. Similarity is an important underlying concept for the above techniques. Depending on the representation mechanism of information items different similarity approaches have been established in the fields of information retrieval and case-based reasoning. However, many times product descriptions consist of both, structured attribute value pairs and free-text descriptions. Therefore, we present a hybrid similarity approach from information retrieval and case-based recommendation systems and enrich it with additional knowledge-based concepts like threshold values and explanations. Furthermore, we implemented our hybrid similarity concept in a service component and give evaluation results for the e-tourism domain.