Uma abordagem de recomendação de tags semânticas para sistemas baseados em tagging

Abstract

Due to the popularization of the Web large amounts of content became available, mainly due to the current ease of content publishing by nontechnical users. This easiness promotes an uncontrollable growth of the total available information, thus creating the need for mechanisms for indexing content to assist users when retrieving information. Because of this information overload, it is infeasible for a user to organize their information without using tools that facilitate the recovery process. One way to accomplish this organization is by a categorization system based on tagging, which helps the process of assigning meaning to any object accessible over the Web. However, since the tags used in the categorizations are freely selected by the user, without any strict control to follow, using this type of system can have various limitations, especially when users need to retrieve an object categorized. This problem gets harder as the users make use of the system, because their vocabulary can become inconsistent due to typing errors, the creation of synonyms and polysemic terms, the introduction of different lexical forms of writing for the same conceptual representation, etc., and, consequently, many relevant resources may be excluded from the search results. Therefore, in this thesis it is proposed an approach for the recommendation of semantic tags that aim to improve the quality of the categorizations made by users looking forward to facilitate the information retrieval, and also to solve the of cold-start problem. The proposal combines three different information sources to generate the recommendation: (i) the textual elements of the Web page, (ii) the folksonomy tags of the page in question, and (iii) the user personomy. Thus, the proposed approach tries to personalize the recommendations based on the user characteristics and interests, as well as to reduce the problems related to the process of tagging. The results for the acceptance of the recommended tags from the execution of some experiments with real users were favorable for the categorization task, which tends to reduce the efforts necessary in the recovery phase of the objects categorized.