Uma estratégia de recomendação associativa de etiquetas usando grafo de contexto em estado de Cold Start

Abstract

The growing production of multimedia content on the web has stimulated studies related to the improvement of the strategies responsible for the organization and recovery of this content in the applications. Among the several strategies, the assignment of a set of tags to an object expressing its content, called the tagging process, has been the subject of recent studies, mainly associative methods that are based on the exploration of co-occurrence patterns of tags. Although associative methods present state of the art results, few investigate the behavior of this type of recommendation in a state in which the object is being inserted into the system and has no previously associated information, characterized as cold start problem. To investigate the behavior of strategies in a cold start, its main motivation is to improve the quality of the recommendations, especially when the strategies are dependent on the information provided by the publishers of the content. This work presents an associative tags recommendation strategy that explores the concept of context graphs through the integration of co-occurrence and tags relevance metrics with the collaborative knowledge of existing relationships between concepts found in Wikipedia articles. With the balance in the valorization of the candidate tags for the recommendation, the proposed strategy obtained improvement in the quality of the re-commendation of the tags in videos that have a median number of tags.