Citation Context

...both supervised and unsupervised methods that have been used in the past with a similar degree of success. Supervised methods generally employ machine learning such as Naive Bayes [9], decision trees =-=[28]-=-, or rule induction [10], using features such as syntactic features, syntactic patterns, and others. On the other side, unsupervised methods, such as e.g. the random walk based system proposed in [19]...

Citation Context

... vocabulary, but instead chooses keyphrases from the text itself. It employs lexical and information retrieval techniques to extract phrases from the document text that are likely to characterize it (=-=Turney, 2000-=-). In this approach, the training data is used to tune the parameters of the extraction algorithm. This paper describes the Kea keyphrase extraction algorithm. It is simple and effective, and performs...

Citation Context

...le learning method (Naive Bayes), or both. In this section, we describe this previous research in more detail. 4.1 GenEx One of the best known programs for keyword extraction is Turney’s GenEx system =-=[22]-=-. GenEx is a rule-based keyphrase extraction system with 12 parameters tuned using a Genetic Algorithm. GenEx has been trained and evaluated on a collection of 652 documents of three different types: ...

Citation Context

... machine learning approach, and use different weightings as input features for the weight assignment algorithm. Similar approaches were shown to work well in some previous work on keyphrase detection =-=[12, 29, 15]-=-. Formally, we consider a training set of labeled instances (x1, l1), . . . , (xn, ln), where xi is a feature vector representing the concept ci, and li is a binary label, indicating whether ci ∈ KC. ...

Citation Context

...ackground corpus is small (or even zero). Some work has been done in extracting keyphrases from technical documents treating keyphrase extraction as a supervised learning problem (Frank et al., 1999; =-=Turney, 2000-=-). The portability of a learned classifier across various unstructured/structured text is not clear, however, and the agreement between classifier and human judges is not high. 1 We would like to have...

Citation Context

...verture/Yahoo! Keyword Selector Tool 2 , and Microsoft adCenter Labs’ Keyword Group Detection 3 ). Previous approaches to the keyword generation problem have exploited the content of either Web pages =-=[10, 20, 26]-=- or search engine results [14]. In this paper, we tackle the problem by making use of search engine query-click logs, an approach that has received limited attention in the literature [4]. Search engi...

Citation Context

... with little or no quality text were difficult to characterize properly and, correspondingly, difficult to organize in a hierarchy. We believe that more advanced techniques for keyword indexing (e.g.,=-=[32, 13]-=- for text or [1, 14] for images) and standard techniques for web indexing (e.g., use of anchor text) can substantially improve the hierarchies. 7. RELATED WORK Several methods have been proposed for b...