National and Kapodistrian University of Athens
Universality of path algebra for recommendation problems
Abstract
dc:descriptionFor many of their daily activities, people interact with online systems that offer an unprecedented variety of products and services. To help users make the best choice, recommender systems come to the rescue. Early recommender systems assumed a simple model that involved two types of entities: users (who receive recommendations) and items (that are recommended). As recent applications became more complex to better accommodate user needs, they now have access to richer, multidimensional data, and can involve several kinds of entities. Currently, existing work in the area provides an abundance of recommendation approaches that differ in their philosophy, algorithms, and overall characteristics that aim to tackle various recommendation use cases. Inevitably, this has resulted in an enormous number of solutions that seemingly address different problems, that involve a lot of replicated effort, and that have unclear differences. The main question of this thesis is whether it is possible to devise a small set of dimensions (i.e., operations) that could concisely describe the vast recommendation space, capturing any recommendation possibility? We argue that recommendations can be abstracted as a graph exploration problem and we describe a graph-theoretic framework with two primary parts: (a) a recommendation graph to model all the elements of an (application) domain from a recommendation perspective, including the subjects of recommendation (that receive recommendations) and its objects (recommended to subjects), as well as relationships among them; (b) a set of path operations to infer new edges, i.e., implicit or unknown relationships, by traversing and combining paths on the graph. These operations allow us to express a wide variety of different recommendations as graph explorations. This approach departs from prior work on recommendation systems, where the graph perspective is mostly missing, and graph-based systems, which are more generic and not targeted to wards recommendations. It provides an abstraction of the recommendation process that reveals a common foundation underlying most recommendation approaches and opens up the door to their efficient optimized implementation over a graph engine. The contribution of the thesis can be summarized as follows: 1. We introduce an elegant data model and a computational model to describe recommendation problems as path problems and to build generic finetuned recommenders. 2. We present a Datalog-like query language that fits the proposed framework. 3. We show how different recommendation strategies can be captured uniformly within our model using the proposed operators. 4. We show examples of novel applications and how they can be represented with our model. 5. We present optimization opportunities in the derivation and execution of recommendations by examining properties of the operators of its computational model. 6. We implement a generic recommender prototype on top of the Neo4j graph system and perform a set of experiments on different recommendation paradigms where all the above benefits arise.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2021
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kyriakidi, Maria-Eleni
- Κυριακίδη, Μαρία-Ελένη
Subjects
dc:subject × 18- Συστήματα συστάσεων
- Εξατομίκευση
- Αναπαράσταση γνώσης
- Μοντέλα γράφων
- Άλγεβρα μονοπατιών
- Βάσεις δεδομένων γράφου
- Recommender systems
- Personalization
- Knowledge representation
- Graph models
- Path algebra
- Graph databases
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Πληροφοριακά συστήματα
- Computer and Information Sciences
- Natural Sciences
- Information Systems
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/50036