Analyse einer dynamischen Sammlung von Zeitungsartikeln mit inhaltsbasierten Methoden
| Autor: | Neumeyer, Markus |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2020 |
| Předmět: |
recommendation
dynamic corpus Empfehlung Zeitungsartikel keyword detection keyword surprise news recommender system NLP diversity Empfehlungssystem beyond accuracy Trend news recommender Trendanalyse news natural language processing content-based Dynamischer Textkorpus Nachrichtenempfehlung Nachrichten Überraschung Diversität serendipity positive Überraschung inhaltsbasiert news article comparison Schlagwortfindung CL glücklicher Zufall Computerlinguistik Vergleich Schlagwort trend analysis |
| DOI: | 10.34726/hss.2020.70883 |
| Popis: | The consumption of news changed throughout the last decades, a huge amount of articles is available at any time in the internet. Consumers therefore need help to find articles that might be relevant for them, as they are not able to scan through all offered articles themselves. This led to the emergence of news recommender systems.The way in which these systems choose articles that might be relevant varies vastly. One kind of methods are the content-based methods, which use only the written content of news articles and build relations between articles for the recommendations based on it. In contrast to collaborative filtering methods, which also use demographic data and previously gathered interests of users.In this work we analyze and compare current state of the art methods for content-based recommendations of news articles.The focus of the comparison will be on two main points. On the one hand is the ability to analyze a dynamic corpus. This includes both the possibility to include new articles to an existing model, as well as finding trends within the found topics or keywords of a model. On the other hand comes the diversity and serendipity of recommendations. Most comparisons of recommender systems put the focus on the accuracy of recommendations,instead this thesis will put the focus on diversity and serendipity to further improve the quality of recommendations. The conclusion of this comparison is, that every method has its strengths and weaknesses. No method could be found that exceeds all other methods in all aspects that were considered. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|