Abstract: Mashups have emerged as a popular technique to compose value-added web services/APIs, to fulfill some complicated business needs. This has increased the number of available mashups over the internet. The increase however, poses a new requirement of organizing and managing these mashups for better understanding and discovery. For this reason, tags have become highly important because they describe items and allows for easy discovery. Most existing tag recommendation methods typically follow a manual process based on controlled vocabulary, or consider tags as words in isolation contained in mashup descriptions. Such methods therefore fail to characterize the diverse functional features of mashups. This work proposes an attention model to automatically recommend mashup tags. Specifically, our proposed model has two levels of attention mechanisms applied at the word- and sentence-levels and subsequently recommend top-N words with highest attention weights as tags. Our model is based on the intuition that not every word in a mashup description is equally relevant in identifying its functional aspects. Therefore, determining the relevant sections involves modeling the interactions of the words, not just their presence in isolation. We demonstrate the effectiveness of our method by conducting extensive experiments on a real-world dataset crawled from www.programmableweb.com. We also compare our method with some baseline tag recommendation methods for verification.
Authors: Kenneth K Fletcher (University of Massachusetts Boston, USA)
Email: richard.anarfi001@umb.edu, kenneth.fletcher@umb.edu