With the development of Web 2.0 and the launch of Web Sites like Flickr, sharingrnand collaboratively annotating images(tagging in Folksonomies) with keywords,rncalled tags, are becoming very popular. Although tagging simplifies resourcernbrowsing and retrieval, it suffers from several issues. Among the issues are redundancyrnand ambiguity. Sometimes a tag which is a very important element willrnbe missed, if the user uploads image without tag.This thesis proposed a hybridrnimage annotation technique which consists of both user assisted(semi-automatic)rnand automatic image annotation strategies. The study mainly focuses on the problemrnof (1) resolving tag word-sense ambiguity(tag-word disambiguation) within arntypical semi-automatic tagging procedure, and (2) Recommending tags of thernnew image automatically, if it is uploaded without tags using tags of previouslyrnuploaded similar images based on the result of tags (or words) co-occurrencernanalysis.rnBoth should rely on effective word-to-context relatedness metrics. Among thernmost effective relatedness metrics are those defined on the basis of a featurernvector representation of the words. In the study comparison of different wordto-rncontext relatedness metrics in terms of effectiveness within finding tags (orrnwords) relatedness process is done. Based on the results of the comparison,rnwe propose to use a metrics derived from a Maximum Likelihood estimator ofrnthe Jensen-Shannon Divergence among feature-count histograms and we showrnthat such a metrics performs(in terms of quality of the output) better than bothrnthe Jensen-Shannon and the Symmetrized Kullback-Leibler divergence betweenrnhistograms. The relative gain in quality within the task of unsupervised cue-wordrnviiirndiscovery and tags co-occurrence analysis by using a synthetic language corpusrnhas been studied.rnIn tags relatedness analysis using co-occurrence information, a word is assignedrnto a specific context chosen among the different ones to which it is related.rnRelatedness to a context is often defined based on the co-occurrence of the targetrnword with other words (context words) in sentences of a specific corpus. Contextrnwords play the role of features for the target. The overall disambiguation processrnor tags co-occurrence analysis can be thought as a classification process. Arnproblem with this approach is that a large number of possible context words canrnreduce the classification performance, both in terms of computational effort andrnin terms of quality of the outcome. Feature selection can improve the processrnin both regards, by reducing the overall feature space to a manageable size withrnhigh information content. In this work, in disambiguation or tags co-occurrencernanalysis, a novel approach using a feature selection based on the Shapley Valuern(SV) â€“ a Coalitional Game Theory related metrics, measuring the importance ofrna component within a coalition is proposed. By including in the feature set onlyrnthe words with highest Shapley Value, tags quality(correctness of tags) and performancernimprovements are obtained. The problem of the exponential complexityrnin the exact SV computation is avoided by an approximate computation based onrnsampling. The study demonstrates the effectiveness of this method and of thernsampling approach results, by using both a synthetic language corpus, a corpusrnprepared from Flickr images database(previously annotated images) and a realrnworld linguistic corpus from Wikipedia English document dump.rnWe showed the extent to which each of the procedures in our approach contributesrnto the overall performance improvements using standard evaluation metrics.